[tutorial] edited hands-on practices (#1899)

* Add handson to ColossalAI.

* Change names of handsons and edit sequence parallel example.

* Edit wrong folder name

* resolve conflict

* delete readme

examples/tutorial/stable_diffusion/ldm/modules/image_degradation/utils_image.py

+import os
+import math
+import random
+import numpy as np
+import torch
+import cv2
+from torchvision.utils import make_grid
+from datetime import datetime
+#import matplotlib.pyplot as plt   # TODO: check with Dominik, also bsrgan.py vs bsrgan_light.py
+
+
+os.environ["KMP_DUPLICATE_LIB_OK"]="TRUE"
+
+
+'''
+# --------------------------------------------
+# Kai Zhang (github: https://github.com/cszn)
+# 03/Mar/2019
+# --------------------------------------------
+# https://github.com/twhui/SRGAN-pyTorch
+# https://github.com/xinntao/BasicSR
+# --------------------------------------------
+'''
+
+
+IMG_EXTENSIONS = ['.jpg', '.JPG', '.jpeg', '.JPEG', '.png', '.PNG', '.ppm', '.PPM', '.bmp', '.BMP', '.tif']
+
+
+def is_image_file(filename):
+    return any(filename.endswith(extension) for extension in IMG_EXTENSIONS)
+
+
+def get_timestamp():
+    return datetime.now().strftime('%y%m%d-%H%M%S')
+
+
+def imshow(x, title=None, cbar=False, figsize=None):
+    plt.figure(figsize=figsize)
+    plt.imshow(np.squeeze(x), interpolation='nearest', cmap='gray')
+    if title:
+        plt.title(title)
+    if cbar:
+        plt.colorbar()
+    plt.show()
+
+
+def surf(Z, cmap='rainbow', figsize=None):
+    plt.figure(figsize=figsize)
+    ax3 = plt.axes(projection='3d')
+
+    w, h = Z.shape[:2]
+    xx = np.arange(0,w,1)
+    yy = np.arange(0,h,1)
+    X, Y = np.meshgrid(xx, yy)
+    ax3.plot_surface(X,Y,Z,cmap=cmap)
+    #ax3.contour(X,Y,Z, zdim='z',offset=-2，cmap=cmap)
+    plt.show()
+
+
+'''
+# --------------------------------------------
+# get image pathes
+# --------------------------------------------
+'''
+
+
+def get_image_paths(dataroot):
+    paths = None  # return None if dataroot is None
+    if dataroot is not None:
+        paths = sorted(_get_paths_from_images(dataroot))
+    return paths
+
+
+def _get_paths_from_images(path):
+    assert os.path.isdir(path), '{:s} is not a valid directory'.format(path)
+    images = []
+    for dirpath, _, fnames in sorted(os.walk(path)):
+        for fname in sorted(fnames):
+            if is_image_file(fname):
+                img_path = os.path.join(dirpath, fname)
+                images.append(img_path)
+    assert images, '{:s} has no valid image file'.format(path)
+    return images
+
+
+'''
+# --------------------------------------------
+# split large images into small images 
+# --------------------------------------------
+'''
+
+
+def patches_from_image(img, p_size=512, p_overlap=64, p_max=800):
+    w, h = img.shape[:2]
+    patches = []
+    if w > p_max and h > p_max:
+        w1 = list(np.arange(0, w-p_size, p_size-p_overlap, dtype=np.int))
+        h1 = list(np.arange(0, h-p_size, p_size-p_overlap, dtype=np.int))
+        w1.append(w-p_size)
+        h1.append(h-p_size)
+#        print(w1)
+#        print(h1)
+        for i in w1:
+            for j in h1:
+                patches.append(img[i:i+p_size, j:j+p_size,:])
+    else:
+        patches.append(img)
+
+    return patches
+
+
+def imssave(imgs, img_path):
+    """
+    imgs: list, N images of size WxHxC
+    """
+    img_name, ext = os.path.splitext(os.path.basename(img_path))
+
+    for i, img in enumerate(imgs):
+        if img.ndim == 3:
+            img = img[:, :, [2, 1, 0]]
+        new_path = os.path.join(os.path.dirname(img_path), img_name+str('_s{:04d}'.format(i))+'.png')
+        cv2.imwrite(new_path, img)
+
+
+def split_imageset(original_dataroot, taget_dataroot, n_channels=3, p_size=800, p_overlap=96, p_max=1000):
+    """
+    split the large images from original_dataroot into small overlapped images with size (p_size)x(p_size),
+    and save them into taget_dataroot; only the images with larger size than (p_max)x(p_max)
+    will be splitted.
+    Args:
+        original_dataroot:
+        taget_dataroot:
+        p_size: size of small images
+        p_overlap: patch size in training is a good choice
+        p_max: images with smaller size than (p_max)x(p_max) keep unchanged.
+    """
+    paths = get_image_paths(original_dataroot)
+    for img_path in paths:
+        # img_name, ext = os.path.splitext(os.path.basename(img_path))
+        img = imread_uint(img_path, n_channels=n_channels)
+        patches = patches_from_image(img, p_size, p_overlap, p_max)
+        imssave(patches, os.path.join(taget_dataroot,os.path.basename(img_path)))
+        #if original_dataroot == taget_dataroot:
+        #del img_path
+
+'''
+# --------------------------------------------
+# makedir
+# --------------------------------------------
+'''
+
+
+def mkdir(path):
+    if not os.path.exists(path):
+        os.makedirs(path)
+
+
+def mkdirs(paths):
+    if isinstance(paths, str):
+        mkdir(paths)
+    else:
+        for path in paths:
+            mkdir(path)
+
+
+def mkdir_and_rename(path):
+    if os.path.exists(path):
+        new_name = path + '_archived_' + get_timestamp()
+        print('Path already exists. Rename it to [{:s}]'.format(new_name))
+        os.rename(path, new_name)
+    os.makedirs(path)
+
+
+'''
+# --------------------------------------------
+# read image from path
+# opencv is fast, but read BGR numpy image
+# --------------------------------------------
+'''
+
+
+# --------------------------------------------
+# get uint8 image of size HxWxn_channles (RGB)
+# --------------------------------------------
+def imread_uint(path, n_channels=3):
+    #  input: path
+    # output: HxWx3(RGB or GGG), or HxWx1 (G)
+    if n_channels == 1:
+        img = cv2.imread(path, 0)  # cv2.IMREAD_GRAYSCALE
+        img = np.expand_dims(img, axis=2)  # HxWx1
+    elif n_channels == 3:
+        img = cv2.imread(path, cv2.IMREAD_UNCHANGED)  # BGR or G
+        if img.ndim == 2:
+            img = cv2.cvtColor(img, cv2.COLOR_GRAY2RGB)  # GGG
+        else:
+            img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)  # RGB
+    return img
+
+
+# --------------------------------------------
+# matlab's imwrite
+# --------------------------------------------
+def imsave(img, img_path):
+    img = np.squeeze(img)
+    if img.ndim == 3:
+        img = img[:, :, [2, 1, 0]]
+    cv2.imwrite(img_path, img)
+
+def imwrite(img, img_path):
+    img = np.squeeze(img)
+    if img.ndim == 3:
+        img = img[:, :, [2, 1, 0]]
+    cv2.imwrite(img_path, img)
+
+
+
+# --------------------------------------------
+# get single image of size HxWxn_channles (BGR)
+# --------------------------------------------
+def read_img(path):
+    # read image by cv2
+    # return: Numpy float32, HWC, BGR, [0,1]
+    img = cv2.imread(path, cv2.IMREAD_UNCHANGED)  # cv2.IMREAD_GRAYSCALE
+    img = img.astype(np.float32) / 255.
+    if img.ndim == 2:
+        img = np.expand_dims(img, axis=2)
+    # some images have 4 channels
+    if img.shape[2] > 3:
+        img = img[:, :, :3]
+    return img
+
+
+'''
+# --------------------------------------------
+# image format conversion
+# --------------------------------------------
+# numpy(single) <--->  numpy(unit)
+# numpy(single) <--->  tensor
+# numpy(unit)   <--->  tensor
+# --------------------------------------------
+'''
+
+
+# --------------------------------------------
+# numpy(single) [0, 1] <--->  numpy(unit)
+# --------------------------------------------
+
+
+def uint2single(img):
+
+    return np.float32(img/255.)
+
+
+def single2uint(img):
+
+    return np.uint8((img.clip(0, 1)*255.).round())
+
+
+def uint162single(img):
+
+    return np.float32(img/65535.)
+
+
+def single2uint16(img):
+
+    return np.uint16((img.clip(0, 1)*65535.).round())
+
+
+# --------------------------------------------
+# numpy(unit) (HxWxC or HxW) <--->  tensor
+# --------------------------------------------
+
+
+# convert uint to 4-dimensional torch tensor
+def uint2tensor4(img):
+    if img.ndim == 2:
+        img = np.expand_dims(img, axis=2)
+    return torch.from_numpy(np.ascontiguousarray(img)).permute(2, 0, 1).float().div(255.).unsqueeze(0)
+
+
+# convert uint to 3-dimensional torch tensor
+def uint2tensor3(img):
+    if img.ndim == 2:
+        img = np.expand_dims(img, axis=2)
+    return torch.from_numpy(np.ascontiguousarray(img)).permute(2, 0, 1).float().div(255.)
+
+
+# convert 2/3/4-dimensional torch tensor to uint
+def tensor2uint(img):
+    img = img.data.squeeze().float().clamp_(0, 1).cpu().numpy()
+    if img.ndim == 3:
+        img = np.transpose(img, (1, 2, 0))
+    return np.uint8((img*255.0).round())
+
+
+# --------------------------------------------
+# numpy(single) (HxWxC) <--->  tensor
+# --------------------------------------------
+
+
+# convert single (HxWxC) to 3-dimensional torch tensor
+def single2tensor3(img):
+    return torch.from_numpy(np.ascontiguousarray(img)).permute(2, 0, 1).float()
+
+
+# convert single (HxWxC) to 4-dimensional torch tensor
+def single2tensor4(img):
+    return torch.from_numpy(np.ascontiguousarray(img)).permute(2, 0, 1).float().unsqueeze(0)
+
+
+# convert torch tensor to single
+def tensor2single(img):
+    img = img.data.squeeze().float().cpu().numpy()
+    if img.ndim == 3:
+        img = np.transpose(img, (1, 2, 0))
+
+    return img
+
+# convert torch tensor to single
+def tensor2single3(img):
+    img = img.data.squeeze().float().cpu().numpy()
+    if img.ndim == 3:
+        img = np.transpose(img, (1, 2, 0))
+    elif img.ndim == 2:
+        img = np.expand_dims(img, axis=2)
+    return img
+
+
+def single2tensor5(img):
+    return torch.from_numpy(np.ascontiguousarray(img)).permute(2, 0, 1, 3).float().unsqueeze(0)
+
+
+def single32tensor5(img):
+    return torch.from_numpy(np.ascontiguousarray(img)).float().unsqueeze(0).unsqueeze(0)
+
+
+def single42tensor4(img):
+    return torch.from_numpy(np.ascontiguousarray(img)).permute(2, 0, 1, 3).float()
+
+
+# from skimage.io import imread, imsave
+def tensor2img(tensor, out_type=np.uint8, min_max=(0, 1)):
+    '''
+    Converts a torch Tensor into an image Numpy array of BGR channel order
+    Input: 4D(B,(3/1),H,W), 3D(C,H,W), or 2D(H,W), any range, RGB channel order
+    Output: 3D(H,W,C) or 2D(H,W), [0,255], np.uint8 (default)
+    '''
+    tensor = tensor.squeeze().float().cpu().clamp_(*min_max)  # squeeze first, then clamp
+    tensor = (tensor - min_max[0]) / (min_max[1] - min_max[0])  # to range [0,1]
+    n_dim = tensor.dim()
+    if n_dim == 4:
+        n_img = len(tensor)
+        img_np = make_grid(tensor, nrow=int(math.sqrt(n_img)), normalize=False).numpy()
+        img_np = np.transpose(img_np[[2, 1, 0], :, :], (1, 2, 0))  # HWC, BGR
+    elif n_dim == 3:
+        img_np = tensor.numpy()
+        img_np = np.transpose(img_np[[2, 1, 0], :, :], (1, 2, 0))  # HWC, BGR
+    elif n_dim == 2:
+        img_np = tensor.numpy()
+    else:
+        raise TypeError(
+            'Only support 4D, 3D and 2D tensor. But received with dimension: {:d}'.format(n_dim))
+    if out_type == np.uint8:
+        img_np = (img_np * 255.0).round()
+        # Important. Unlike matlab, numpy.unit8() WILL NOT round by default.
+    return img_np.astype(out_type)
+
+
+'''
+# --------------------------------------------
+# Augmentation, flipe and/or rotate
+# --------------------------------------------
+# The following two are enough.
+# (1) augmet_img: numpy image of WxHxC or WxH
+# (2) augment_img_tensor4: tensor image 1xCxWxH
+# --------------------------------------------
+'''
+
+
+def augment_img(img, mode=0):
+    '''Kai Zhang (github: https://github.com/cszn)
+    '''
+    if mode == 0:
+        return img
+    elif mode == 1:
+        return np.flipud(np.rot90(img))
+    elif mode == 2:
+        return np.flipud(img)
+    elif mode == 3:
+        return np.rot90(img, k=3)
+    elif mode == 4:
+        return np.flipud(np.rot90(img, k=2))
+    elif mode == 5:
+        return np.rot90(img)
+    elif mode == 6:
+        return np.rot90(img, k=2)
+    elif mode == 7:
+        return np.flipud(np.rot90(img, k=3))
+
+
+def augment_img_tensor4(img, mode=0):
+    '''Kai Zhang (github: https://github.com/cszn)
+    '''
+    if mode == 0:
+        return img
+    elif mode == 1:
+        return img.rot90(1, [2, 3]).flip([2])
+    elif mode == 2:
+        return img.flip([2])
+    elif mode == 3:
+        return img.rot90(3, [2, 3])
+    elif mode == 4:
+        return img.rot90(2, [2, 3]).flip([2])
+    elif mode == 5:
+        return img.rot90(1, [2, 3])
+    elif mode == 6:
+        return img.rot90(2, [2, 3])
+    elif mode == 7:
+        return img.rot90(3, [2, 3]).flip([2])
+
+
+def augment_img_tensor(img, mode=0):
+    '''Kai Zhang (github: https://github.com/cszn)
+    '''
+    img_size = img.size()
+    img_np = img.data.cpu().numpy()
+    if len(img_size) == 3:
+        img_np = np.transpose(img_np, (1, 2, 0))
+    elif len(img_size) == 4:
+        img_np = np.transpose(img_np, (2, 3, 1, 0))
+    img_np = augment_img(img_np, mode=mode)
+    img_tensor = torch.from_numpy(np.ascontiguousarray(img_np))
+    if len(img_size) == 3:
+        img_tensor = img_tensor.permute(2, 0, 1)
+    elif len(img_size) == 4:
+        img_tensor = img_tensor.permute(3, 2, 0, 1)
+
+    return img_tensor.type_as(img)
+
+
+def augment_img_np3(img, mode=0):
+    if mode == 0:
+        return img
+    elif mode == 1:
+        return img.transpose(1, 0, 2)
+    elif mode == 2:
+        return img[::-1, :, :]
+    elif mode == 3:
+        img = img[::-1, :, :]
+        img = img.transpose(1, 0, 2)
+        return img
+    elif mode == 4:
+        return img[:, ::-1, :]
+    elif mode == 5:
+        img = img[:, ::-1, :]
+        img = img.transpose(1, 0, 2)
+        return img
+    elif mode == 6:
+        img = img[:, ::-1, :]
+        img = img[::-1, :, :]
+        return img
+    elif mode == 7:
+        img = img[:, ::-1, :]
+        img = img[::-1, :, :]
+        img = img.transpose(1, 0, 2)
+        return img
+
+
+def augment_imgs(img_list, hflip=True, rot=True):
+    # horizontal flip OR rotate
+    hflip = hflip and random.random() < 0.5
+    vflip = rot and random.random() < 0.5
+    rot90 = rot and random.random() < 0.5
+
+    def _augment(img):
+        if hflip:
+            img = img[:, ::-1, :]
+        if vflip:
+            img = img[::-1, :, :]
+        if rot90:
+            img = img.transpose(1, 0, 2)
+        return img
+
+    return [_augment(img) for img in img_list]
+
+
+'''
+# --------------------------------------------
+# modcrop and shave
+# --------------------------------------------
+'''
+
+
+def modcrop(img_in, scale):
+    # img_in: Numpy, HWC or HW
+    img = np.copy(img_in)
+    if img.ndim == 2:
+        H, W = img.shape
+        H_r, W_r = H % scale, W % scale
+        img = img[:H - H_r, :W - W_r]
+    elif img.ndim == 3:
+        H, W, C = img.shape
+        H_r, W_r = H % scale, W % scale
+        img = img[:H - H_r, :W - W_r, :]
+    else:
+        raise ValueError('Wrong img ndim: [{:d}].'.format(img.ndim))
+    return img
+
+
+def shave(img_in, border=0):
+    # img_in: Numpy, HWC or HW
+    img = np.copy(img_in)
+    h, w = img.shape[:2]
+    img = img[border:h-border, border:w-border]
+    return img
+
+
+'''
+# --------------------------------------------
+# image processing process on numpy image
+# channel_convert(in_c, tar_type, img_list):
+# rgb2ycbcr(img, only_y=True):
+# bgr2ycbcr(img, only_y=True):
+# ycbcr2rgb(img):
+# --------------------------------------------
+'''
+
+
+def rgb2ycbcr(img, only_y=True):
+    '''same as matlab rgb2ycbcr
+    only_y: only return Y channel
+    Input:
+        uint8, [0, 255]
+        float, [0, 1]
+    '''
+    in_img_type = img.dtype
+    img.astype(np.float32)
+    if in_img_type != np.uint8:
+        img *= 255.
+    # convert
+    if only_y:
+        rlt = np.dot(img, [65.481, 128.553, 24.966]) / 255.0 + 16.0
+    else:
+        rlt = np.matmul(img, [[65.481, -37.797, 112.0], [128.553, -74.203, -93.786],
+                              [24.966, 112.0, -18.214]]) / 255.0 + [16, 128, 128]
+    if in_img_type == np.uint8:
+        rlt = rlt.round()
+    else:
+        rlt /= 255.
+    return rlt.astype(in_img_type)
+
+
+def ycbcr2rgb(img):
+    '''same as matlab ycbcr2rgb
+    Input:
+        uint8, [0, 255]
+        float, [0, 1]
+    '''
+    in_img_type = img.dtype
+    img.astype(np.float32)
+    if in_img_type != np.uint8:
+        img *= 255.
+    # convert
+    rlt = np.matmul(img, [[0.00456621, 0.00456621, 0.00456621], [0, -0.00153632, 0.00791071],
+                          [0.00625893, -0.00318811, 0]]) * 255.0 + [-222.921, 135.576, -276.836]
+    if in_img_type == np.uint8:
+        rlt = rlt.round()
+    else:
+        rlt /= 255.
+    return rlt.astype(in_img_type)
+
+
+def bgr2ycbcr(img, only_y=True):
+    '''bgr version of rgb2ycbcr
+    only_y: only return Y channel
+    Input:
+        uint8, [0, 255]
+        float, [0, 1]
+    '''
+    in_img_type = img.dtype
+    img.astype(np.float32)
+    if in_img_type != np.uint8:
+        img *= 255.
+    # convert
+    if only_y:
+        rlt = np.dot(img, [24.966, 128.553, 65.481]) / 255.0 + 16.0
+    else:
+        rlt = np.matmul(img, [[24.966, 112.0, -18.214], [128.553, -74.203, -93.786],
+                              [65.481, -37.797, 112.0]]) / 255.0 + [16, 128, 128]
+    if in_img_type == np.uint8:
+        rlt = rlt.round()
+    else:
+        rlt /= 255.
+    return rlt.astype(in_img_type)
+
+
+def channel_convert(in_c, tar_type, img_list):
+    # conversion among BGR, gray and y
+    if in_c == 3 and tar_type == 'gray':  # BGR to gray
+        gray_list = [cv2.cvtColor(img, cv2.COLOR_BGR2GRAY) for img in img_list]
+        return [np.expand_dims(img, axis=2) for img in gray_list]
+    elif in_c == 3 and tar_type == 'y':  # BGR to y
+        y_list = [bgr2ycbcr(img, only_y=True) for img in img_list]
+        return [np.expand_dims(img, axis=2) for img in y_list]
+    elif in_c == 1 and tar_type == 'RGB':  # gray/y to BGR
+        return [cv2.cvtColor(img, cv2.COLOR_GRAY2BGR) for img in img_list]
+    else:
+        return img_list
+
+
+'''
+# --------------------------------------------
+# metric, PSNR and SSIM
+# --------------------------------------------
+'''
+
+
+# --------------------------------------------
+# PSNR
+# --------------------------------------------
+def calculate_psnr(img1, img2, border=0):
+    # img1 and img2 have range [0, 255]
+    #img1 = img1.squeeze()
+    #img2 = img2.squeeze()
+    if not img1.shape == img2.shape:
+        raise ValueError('Input images must have the same dimensions.')
+    h, w = img1.shape[:2]
+    img1 = img1[border:h-border, border:w-border]
+    img2 = img2[border:h-border, border:w-border]
+
+    img1 = img1.astype(np.float64)
+    img2 = img2.astype(np.float64)
+    mse = np.mean((img1 - img2)**2)
+    if mse == 0:
+        return float('inf')
+    return 20 * math.log10(255.0 / math.sqrt(mse))
+
+
+# --------------------------------------------
+# SSIM
+# --------------------------------------------
+def calculate_ssim(img1, img2, border=0):
+    '''calculate SSIM
+    the same outputs as MATLAB's
+    img1, img2: [0, 255]
+    '''
+    #img1 = img1.squeeze()
+    #img2 = img2.squeeze()
+    if not img1.shape == img2.shape:
+        raise ValueError('Input images must have the same dimensions.')
+    h, w = img1.shape[:2]
+    img1 = img1[border:h-border, border:w-border]
+    img2 = img2[border:h-border, border:w-border]
+
+    if img1.ndim == 2:
+        return ssim(img1, img2)
+    elif img1.ndim == 3:
+        if img1.shape[2] == 3:
+            ssims = []
+            for i in range(3):
+                ssims.append(ssim(img1[:,:,i], img2[:,:,i]))
+            return np.array(ssims).mean()
+        elif img1.shape[2] == 1:
+            return ssim(np.squeeze(img1), np.squeeze(img2))
+    else:
+        raise ValueError('Wrong input image dimensions.')
+
+
+def ssim(img1, img2):
+    C1 = (0.01 * 255)**2
+    C2 = (0.03 * 255)**2
+
+    img1 = img1.astype(np.float64)
+    img2 = img2.astype(np.float64)
+    kernel = cv2.getGaussianKernel(11, 1.5)
+    window = np.outer(kernel, kernel.transpose())
+
+    mu1 = cv2.filter2D(img1, -1, window)[5:-5, 5:-5]  # valid
+    mu2 = cv2.filter2D(img2, -1, window)[5:-5, 5:-5]
+    mu1_sq = mu1**2
+    mu2_sq = mu2**2
+    mu1_mu2 = mu1 * mu2
+    sigma1_sq = cv2.filter2D(img1**2, -1, window)[5:-5, 5:-5] - mu1_sq
+    sigma2_sq = cv2.filter2D(img2**2, -1, window)[5:-5, 5:-5] - mu2_sq
+    sigma12 = cv2.filter2D(img1 * img2, -1, window)[5:-5, 5:-5] - mu1_mu2
+
+    ssim_map = ((2 * mu1_mu2 + C1) * (2 * sigma12 + C2)) / ((mu1_sq + mu2_sq + C1) *
+                                                            (sigma1_sq + sigma2_sq + C2))
+    return ssim_map.mean()
+
+
+'''
+# --------------------------------------------
+# matlab's bicubic imresize (numpy and torch) [0, 1]
+# --------------------------------------------
+'''
+
+
+# matlab 'imresize' function, now only support 'bicubic'
+def cubic(x):
+    absx = torch.abs(x)
+    absx2 = absx**2
+    absx3 = absx**3
+    return (1.5*absx3 - 2.5*absx2 + 1) * ((absx <= 1).type_as(absx)) + \
+        (-0.5*absx3 + 2.5*absx2 - 4*absx + 2) * (((absx > 1)*(absx <= 2)).type_as(absx))
+
+
+def calculate_weights_indices(in_length, out_length, scale, kernel, kernel_width, antialiasing):
+    if (scale < 1) and (antialiasing):
+        # Use a modified kernel to simultaneously interpolate and antialias- larger kernel width
+        kernel_width = kernel_width / scale
+
+    # Output-space coordinates
+    x = torch.linspace(1, out_length, out_length)
+
+    # Input-space coordinates. Calculate the inverse mapping such that 0.5
+    # in output space maps to 0.5 in input space, and 0.5+scale in output
+    # space maps to 1.5 in input space.
+    u = x / scale + 0.5 * (1 - 1 / scale)
+
+    # What is the left-most pixel that can be involved in the computation?
+    left = torch.floor(u - kernel_width / 2)
+
+    # What is the maximum number of pixels that can be involved in the
+    # computation?  Note: it's OK to use an extra pixel here; if the
+    # corresponding weights are all zero, it will be eliminated at the end
+    # of this function.
+    P = math.ceil(kernel_width) + 2
+
+    # The indices of the input pixels involved in computing the k-th output
+    # pixel are in row k of the indices matrix.
+    indices = left.view(out_length, 1).expand(out_length, P) + torch.linspace(0, P - 1, P).view(
+        1, P).expand(out_length, P)
+
+    # The weights used to compute the k-th output pixel are in row k of the
+    # weights matrix.
+    distance_to_center = u.view(out_length, 1).expand(out_length, P) - indices
+    # apply cubic kernel
+    if (scale < 1) and (antialiasing):
+        weights = scale * cubic(distance_to_center * scale)
+    else:
+        weights = cubic(distance_to_center)
+    # Normalize the weights matrix so that each row sums to 1.
+    weights_sum = torch.sum(weights, 1).view(out_length, 1)
+    weights = weights / weights_sum.expand(out_length, P)
+
+    # If a column in weights is all zero, get rid of it. only consider the first and last column.
+    weights_zero_tmp = torch.sum((weights == 0), 0)
+    if not math.isclose(weights_zero_tmp[0], 0, rel_tol=1e-6):
+        indices = indices.narrow(1, 1, P - 2)
+        weights = weights.narrow(1, 1, P - 2)
+    if not math.isclose(weights_zero_tmp[-1], 0, rel_tol=1e-6):
+        indices = indices.narrow(1, 0, P - 2)
+        weights = weights.narrow(1, 0, P - 2)
+    weights = weights.contiguous()
+    indices = indices.contiguous()
+    sym_len_s = -indices.min() + 1
+    sym_len_e = indices.max() - in_length
+    indices = indices + sym_len_s - 1
+    return weights, indices, int(sym_len_s), int(sym_len_e)
+
+
+# --------------------------------------------
+# imresize for tensor image [0, 1]
+# --------------------------------------------
+def imresize(img, scale, antialiasing=True):
+    # Now the scale should be the same for H and W
+    # input: img: pytorch tensor, CHW or HW [0,1]
+    # output: CHW or HW [0,1] w/o round
+    need_squeeze = True if img.dim() == 2 else False
+    if need_squeeze:
+        img.unsqueeze_(0)
+    in_C, in_H, in_W = img.size()
+    out_C, out_H, out_W = in_C, math.ceil(in_H * scale), math.ceil(in_W * scale)
+    kernel_width = 4
+    kernel = 'cubic'
+
+    # Return the desired dimension order for performing the resize.  The
+    # strategy is to perform the resize first along the dimension with the
+    # smallest scale factor.
+    # Now we do not support this.
+
+    # get weights and indices
+    weights_H, indices_H, sym_len_Hs, sym_len_He = calculate_weights_indices(
+        in_H, out_H, scale, kernel, kernel_width, antialiasing)
+    weights_W, indices_W, sym_len_Ws, sym_len_We = calculate_weights_indices(
+        in_W, out_W, scale, kernel, kernel_width, antialiasing)
+    # process H dimension
+    # symmetric copying
+    img_aug = torch.FloatTensor(in_C, in_H + sym_len_Hs + sym_len_He, in_W)
+    img_aug.narrow(1, sym_len_Hs, in_H).copy_(img)
+
+    sym_patch = img[:, :sym_len_Hs, :]
+    inv_idx = torch.arange(sym_patch.size(1) - 1, -1, -1).long()
+    sym_patch_inv = sym_patch.index_select(1, inv_idx)
+    img_aug.narrow(1, 0, sym_len_Hs).copy_(sym_patch_inv)
+
+    sym_patch = img[:, -sym_len_He:, :]
+    inv_idx = torch.arange(sym_patch.size(1) - 1, -1, -1).long()
+    sym_patch_inv = sym_patch.index_select(1, inv_idx)
+    img_aug.narrow(1, sym_len_Hs + in_H, sym_len_He).copy_(sym_patch_inv)
+
+    out_1 = torch.FloatTensor(in_C, out_H, in_W)
+    kernel_width = weights_H.size(1)
+    for i in range(out_H):
+        idx = int(indices_H[i][0])
+        for j in range(out_C):
+            out_1[j, i, :] = img_aug[j, idx:idx + kernel_width, :].transpose(0, 1).mv(weights_H[i])
+
+    # process W dimension
+    # symmetric copying
+    out_1_aug = torch.FloatTensor(in_C, out_H, in_W + sym_len_Ws + sym_len_We)
+    out_1_aug.narrow(2, sym_len_Ws, in_W).copy_(out_1)
+
+    sym_patch = out_1[:, :, :sym_len_Ws]
+    inv_idx = torch.arange(sym_patch.size(2) - 1, -1, -1).long()
+    sym_patch_inv = sym_patch.index_select(2, inv_idx)
+    out_1_aug.narrow(2, 0, sym_len_Ws).copy_(sym_patch_inv)
+
+    sym_patch = out_1[:, :, -sym_len_We:]
+    inv_idx = torch.arange(sym_patch.size(2) - 1, -1, -1).long()
+    sym_patch_inv = sym_patch.index_select(2, inv_idx)
+    out_1_aug.narrow(2, sym_len_Ws + in_W, sym_len_We).copy_(sym_patch_inv)
+
+    out_2 = torch.FloatTensor(in_C, out_H, out_W)
+    kernel_width = weights_W.size(1)
+    for i in range(out_W):
+        idx = int(indices_W[i][0])
+        for j in range(out_C):
+            out_2[j, :, i] = out_1_aug[j, :, idx:idx + kernel_width].mv(weights_W[i])
+    if need_squeeze:
+        out_2.squeeze_()
+    return out_2
+
+
+# --------------------------------------------
+# imresize for numpy image [0, 1]
+# --------------------------------------------
+def imresize_np(img, scale, antialiasing=True):
+    # Now the scale should be the same for H and W
+    # input: img: Numpy, HWC or HW [0,1]
+    # output: HWC or HW [0,1] w/o round
+    img = torch.from_numpy(img)
+    need_squeeze = True if img.dim() == 2 else False
+    if need_squeeze:
+        img.unsqueeze_(2)
+
+    in_H, in_W, in_C = img.size()
+    out_C, out_H, out_W = in_C, math.ceil(in_H * scale), math.ceil(in_W * scale)
+    kernel_width = 4
+    kernel = 'cubic'
+
+    # Return the desired dimension order for performing the resize.  The
+    # strategy is to perform the resize first along the dimension with the
+    # smallest scale factor.
+    # Now we do not support this.
+
+    # get weights and indices
+    weights_H, indices_H, sym_len_Hs, sym_len_He = calculate_weights_indices(
+        in_H, out_H, scale, kernel, kernel_width, antialiasing)
+    weights_W, indices_W, sym_len_Ws, sym_len_We = calculate_weights_indices(
+        in_W, out_W, scale, kernel, kernel_width, antialiasing)
+    # process H dimension
+    # symmetric copying
+    img_aug = torch.FloatTensor(in_H + sym_len_Hs + sym_len_He, in_W, in_C)
+    img_aug.narrow(0, sym_len_Hs, in_H).copy_(img)
+
+    sym_patch = img[:sym_len_Hs, :, :]
+    inv_idx = torch.arange(sym_patch.size(0) - 1, -1, -1).long()
+    sym_patch_inv = sym_patch.index_select(0, inv_idx)
+    img_aug.narrow(0, 0, sym_len_Hs).copy_(sym_patch_inv)
+
+    sym_patch = img[-sym_len_He:, :, :]
+    inv_idx = torch.arange(sym_patch.size(0) - 1, -1, -1).long()
+    sym_patch_inv = sym_patch.index_select(0, inv_idx)
+    img_aug.narrow(0, sym_len_Hs + in_H, sym_len_He).copy_(sym_patch_inv)
+
+    out_1 = torch.FloatTensor(out_H, in_W, in_C)
+    kernel_width = weights_H.size(1)
+    for i in range(out_H):
+        idx = int(indices_H[i][0])
+        for j in range(out_C):
+            out_1[i, :, j] = img_aug[idx:idx + kernel_width, :, j].transpose(0, 1).mv(weights_H[i])
+
+    # process W dimension
+    # symmetric copying
+    out_1_aug = torch.FloatTensor(out_H, in_W + sym_len_Ws + sym_len_We, in_C)
+    out_1_aug.narrow(1, sym_len_Ws, in_W).copy_(out_1)
+
+    sym_patch = out_1[:, :sym_len_Ws, :]
+    inv_idx = torch.arange(sym_patch.size(1) - 1, -1, -1).long()
+    sym_patch_inv = sym_patch.index_select(1, inv_idx)
+    out_1_aug.narrow(1, 0, sym_len_Ws).copy_(sym_patch_inv)
+
+    sym_patch = out_1[:, -sym_len_We:, :]
+    inv_idx = torch.arange(sym_patch.size(1) - 1, -1, -1).long()
+    sym_patch_inv = sym_patch.index_select(1, inv_idx)
+    out_1_aug.narrow(1, sym_len_Ws + in_W, sym_len_We).copy_(sym_patch_inv)
+
+    out_2 = torch.FloatTensor(out_H, out_W, in_C)
+    kernel_width = weights_W.size(1)
+    for i in range(out_W):
+        idx = int(indices_W[i][0])
+        for j in range(out_C):
+            out_2[:, i, j] = out_1_aug[:, idx:idx + kernel_width, j].mv(weights_W[i])
+    if need_squeeze:
+        out_2.squeeze_()
+
+    return out_2.numpy()
+
+
+if __name__ == '__main__':
+    print('---')
+#    img = imread_uint('test.bmp', 3)
+#    img = uint2single(img)
+#    img_bicubic = imresize_np(img, 1/4)
\ No newline at end of file

examples/tutorial/stable_diffusion/ldm/modules/losses/__init__.py

+from ldm.modules.losses.contperceptual import LPIPSWithDiscriminator
\ No newline at end of file

examples/tutorial/stable_diffusion/ldm/modules/losses/contperceptual.py

+import torch
+import torch.nn as nn
+
+from taming.modules.losses.vqperceptual import *  # TODO: taming dependency yes/no?
+
+
+class LPIPSWithDiscriminator(nn.Module):
+    def __init__(self, disc_start, logvar_init=0.0, kl_weight=1.0, pixelloss_weight=1.0,
+                 disc_num_layers=3, disc_in_channels=3, disc_factor=1.0, disc_weight=1.0,
+                 perceptual_weight=1.0, use_actnorm=False, disc_conditional=False,
+                 disc_loss="hinge"):
+
+        super().__init__()
+        assert disc_loss in ["hinge", "vanilla"]
+        self.kl_weight = kl_weight
+        self.pixel_weight = pixelloss_weight
+        self.perceptual_loss = LPIPS().eval()
+        self.perceptual_weight = perceptual_weight
+        # output log variance
+        self.logvar = nn.Parameter(torch.ones(size=()) * logvar_init)
+
+        self.discriminator = NLayerDiscriminator(input_nc=disc_in_channels,
+                                                 n_layers=disc_num_layers,
+                                                 use_actnorm=use_actnorm
+                                                 ).apply(weights_init)
+        self.discriminator_iter_start = disc_start
+        self.disc_loss = hinge_d_loss if disc_loss == "hinge" else vanilla_d_loss
+        self.disc_factor = disc_factor
+        self.discriminator_weight = disc_weight
+        self.disc_conditional = disc_conditional
+
+    def calculate_adaptive_weight(self, nll_loss, g_loss, last_layer=None):
+        if last_layer is not None:
+            nll_grads = torch.autograd.grad(nll_loss, last_layer, retain_graph=True)[0]
+            g_grads = torch.autograd.grad(g_loss, last_layer, retain_graph=True)[0]
+        else:
+            nll_grads = torch.autograd.grad(nll_loss, self.last_layer[0], retain_graph=True)[0]
+            g_grads = torch.autograd.grad(g_loss, self.last_layer[0], retain_graph=True)[0]
+
+        d_weight = torch.norm(nll_grads) / (torch.norm(g_grads) + 1e-4)
+        d_weight = torch.clamp(d_weight, 0.0, 1e4).detach()
+        d_weight = d_weight * self.discriminator_weight
+        return d_weight
+
+    def forward(self, inputs, reconstructions, posteriors, optimizer_idx,
+                global_step, last_layer=None, cond=None, split="train",
+                weights=None):
+        rec_loss = torch.abs(inputs.contiguous() - reconstructions.contiguous())
+        if self.perceptual_weight > 0:
+            p_loss = self.perceptual_loss(inputs.contiguous(), reconstructions.contiguous())
+            rec_loss = rec_loss + self.perceptual_weight * p_loss
+
+        nll_loss = rec_loss / torch.exp(self.logvar) + self.logvar
+        weighted_nll_loss = nll_loss
+        if weights is not None:
+            weighted_nll_loss = weights*nll_loss
+        weighted_nll_loss = torch.sum(weighted_nll_loss) / weighted_nll_loss.shape[0]
+        nll_loss = torch.sum(nll_loss) / nll_loss.shape[0]
+        kl_loss = posteriors.kl()
+        kl_loss = torch.sum(kl_loss) / kl_loss.shape[0]
+
+        # now the GAN part
+        if optimizer_idx == 0:
+            # generator update
+            if cond is None:
+                assert not self.disc_conditional
+                logits_fake = self.discriminator(reconstructions.contiguous())
+            else:
+                assert self.disc_conditional
+                logits_fake = self.discriminator(torch.cat((reconstructions.contiguous(), cond), dim=1))
+            g_loss = -torch.mean(logits_fake)
+
+            if self.disc_factor > 0.0:
+                try:
+                    d_weight = self.calculate_adaptive_weight(nll_loss, g_loss, last_layer=last_layer)
+                except RuntimeError:
+                    assert not self.training
+                    d_weight = torch.tensor(0.0)
+            else:
+                d_weight = torch.tensor(0.0)
+
+            disc_factor = adopt_weight(self.disc_factor, global_step, threshold=self.discriminator_iter_start)
+            loss = weighted_nll_loss + self.kl_weight * kl_loss + d_weight * disc_factor * g_loss
+
+            log = {"{}/total_loss".format(split): loss.clone().detach().mean(), "{}/logvar".format(split): self.logvar.detach(),
+                   "{}/kl_loss".format(split): kl_loss.detach().mean(), "{}/nll_loss".format(split): nll_loss.detach().mean(),
+                   "{}/rec_loss".format(split): rec_loss.detach().mean(),
+                   "{}/d_weight".format(split): d_weight.detach(),
+                   "{}/disc_factor".format(split): torch.tensor(disc_factor),
+                   "{}/g_loss".format(split): g_loss.detach().mean(),
+                   }
+            return loss, log
+
+        if optimizer_idx == 1:
+            # second pass for discriminator update
+            if cond is None:
+                logits_real = self.discriminator(inputs.contiguous().detach())
+                logits_fake = self.discriminator(reconstructions.contiguous().detach())
+            else:
+                logits_real = self.discriminator(torch.cat((inputs.contiguous().detach(), cond), dim=1))
+                logits_fake = self.discriminator(torch.cat((reconstructions.contiguous().detach(), cond), dim=1))
+
+            disc_factor = adopt_weight(self.disc_factor, global_step, threshold=self.discriminator_iter_start)
+            d_loss = disc_factor * self.disc_loss(logits_real, logits_fake)
+
+            log = {"{}/disc_loss".format(split): d_loss.clone().detach().mean(),
+                   "{}/logits_real".format(split): logits_real.detach().mean(),
+                   "{}/logits_fake".format(split): logits_fake.detach().mean()
+                   }
+            return d_loss, log
+

examples/tutorial/stable_diffusion/ldm/modules/losses/vqperceptual.py

+import torch
+from torch import nn
+import torch.nn.functional as F
+from einops import repeat
+
+from taming.modules.discriminator.model import NLayerDiscriminator, weights_init
+from taming.modules.losses.lpips import LPIPS
+from taming.modules.losses.vqperceptual import hinge_d_loss, vanilla_d_loss
+
+
+def hinge_d_loss_with_exemplar_weights(logits_real, logits_fake, weights):
+    assert weights.shape[0] == logits_real.shape[0] == logits_fake.shape[0]
+    loss_real = torch.mean(F.relu(1. - logits_real), dim=[1,2,3])
+    loss_fake = torch.mean(F.relu(1. + logits_fake), dim=[1,2,3])
+    loss_real = (weights * loss_real).sum() / weights.sum()
+    loss_fake = (weights * loss_fake).sum() / weights.sum()
+    d_loss = 0.5 * (loss_real + loss_fake)
+    return d_loss
+
+def adopt_weight(weight, global_step, threshold=0, value=0.):
+    if global_step < threshold:
+        weight = value
+    return weight
+
+
+def measure_perplexity(predicted_indices, n_embed):
+    # src: https://github.com/karpathy/deep-vector-quantization/blob/main/model.py
+    # eval cluster perplexity. when perplexity == num_embeddings then all clusters are used exactly equally
+    encodings = F.one_hot(predicted_indices, n_embed).float().reshape(-1, n_embed)
+    avg_probs = encodings.mean(0)
+    perplexity = (-(avg_probs * torch.log(avg_probs + 1e-10)).sum()).exp()
+    cluster_use = torch.sum(avg_probs > 0)
+    return perplexity, cluster_use
+
+def l1(x, y):
+    return torch.abs(x-y)
+
+
+def l2(x, y):
+    return torch.pow((x-y), 2)
+
+
+class VQLPIPSWithDiscriminator(nn.Module):
+    def __init__(self, disc_start, codebook_weight=1.0, pixelloss_weight=1.0,
+                 disc_num_layers=3, disc_in_channels=3, disc_factor=1.0, disc_weight=1.0,
+                 perceptual_weight=1.0, use_actnorm=False, disc_conditional=False,
+                 disc_ndf=64, disc_loss="hinge", n_classes=None, perceptual_loss="lpips",
+                 pixel_loss="l1"):
+        super().__init__()
+        assert disc_loss in ["hinge", "vanilla"]
+        assert perceptual_loss in ["lpips", "clips", "dists"]
+        assert pixel_loss in ["l1", "l2"]
+        self.codebook_weight = codebook_weight
+        self.pixel_weight = pixelloss_weight
+        if perceptual_loss == "lpips":
+            print(f"{self.__class__.__name__}: Running with LPIPS.")
+            self.perceptual_loss = LPIPS().eval()
+        else:
+            raise ValueError(f"Unknown perceptual loss: >> {perceptual_loss} <<")
+        self.perceptual_weight = perceptual_weight
+
+        if pixel_loss == "l1":
+            self.pixel_loss = l1
+        else:
+            self.pixel_loss = l2
+
+        self.discriminator = NLayerDiscriminator(input_nc=disc_in_channels,
+                                                 n_layers=disc_num_layers,
+                                                 use_actnorm=use_actnorm,
+                                                 ndf=disc_ndf
+                                                 ).apply(weights_init)
+        self.discriminator_iter_start = disc_start
+        if disc_loss == "hinge":
+            self.disc_loss = hinge_d_loss
+        elif disc_loss == "vanilla":
+            self.disc_loss = vanilla_d_loss
+        else:
+            raise ValueError(f"Unknown GAN loss '{disc_loss}'.")
+        print(f"VQLPIPSWithDiscriminator running with {disc_loss} loss.")
+        self.disc_factor = disc_factor
+        self.discriminator_weight = disc_weight
+        self.disc_conditional = disc_conditional
+        self.n_classes = n_classes
+
+    def calculate_adaptive_weight(self, nll_loss, g_loss, last_layer=None):
+        if last_layer is not None:
+            nll_grads = torch.autograd.grad(nll_loss, last_layer, retain_graph=True)[0]
+            g_grads = torch.autograd.grad(g_loss, last_layer, retain_graph=True)[0]
+        else:
+            nll_grads = torch.autograd.grad(nll_loss, self.last_layer[0], retain_graph=True)[0]
+            g_grads = torch.autograd.grad(g_loss, self.last_layer[0], retain_graph=True)[0]
+
+        d_weight = torch.norm(nll_grads) / (torch.norm(g_grads) + 1e-4)
+        d_weight = torch.clamp(d_weight, 0.0, 1e4).detach()
+        d_weight = d_weight * self.discriminator_weight
+        return d_weight
+
+    def forward(self, codebook_loss, inputs, reconstructions, optimizer_idx,
+                global_step, last_layer=None, cond=None, split="train", predicted_indices=None):
+        if not exists(codebook_loss):
+            codebook_loss = torch.tensor([0.]).to(inputs.device)
+        #rec_loss = torch.abs(inputs.contiguous() - reconstructions.contiguous())
+        rec_loss = self.pixel_loss(inputs.contiguous(), reconstructions.contiguous())
+        if self.perceptual_weight > 0:
+            p_loss = self.perceptual_loss(inputs.contiguous(), reconstructions.contiguous())
+            rec_loss = rec_loss + self.perceptual_weight * p_loss
+        else:
+            p_loss = torch.tensor([0.0])
+
+        nll_loss = rec_loss
+        #nll_loss = torch.sum(nll_loss) / nll_loss.shape[0]
+        nll_loss = torch.mean(nll_loss)
+
+        # now the GAN part
+        if optimizer_idx == 0:
+            # generator update
+            if cond is None:
+                assert not self.disc_conditional
+                logits_fake = self.discriminator(reconstructions.contiguous())
+            else:
+                assert self.disc_conditional
+                logits_fake = self.discriminator(torch.cat((reconstructions.contiguous(), cond), dim=1))
+            g_loss = -torch.mean(logits_fake)
+
+            try:
+                d_weight = self.calculate_adaptive_weight(nll_loss, g_loss, last_layer=last_layer)
+            except RuntimeError:
+                assert not self.training
+                d_weight = torch.tensor(0.0)
+
+            disc_factor = adopt_weight(self.disc_factor, global_step, threshold=self.discriminator_iter_start)
+            loss = nll_loss + d_weight * disc_factor * g_loss + self.codebook_weight * codebook_loss.mean()
+
+            log = {"{}/total_loss".format(split): loss.clone().detach().mean(),
+                   "{}/quant_loss".format(split): codebook_loss.detach().mean(),
+                   "{}/nll_loss".format(split): nll_loss.detach().mean(),
+                   "{}/rec_loss".format(split): rec_loss.detach().mean(),
+                   "{}/p_loss".format(split): p_loss.detach().mean(),
+                   "{}/d_weight".format(split): d_weight.detach(),
+                   "{}/disc_factor".format(split): torch.tensor(disc_factor),
+                   "{}/g_loss".format(split): g_loss.detach().mean(),
+                   }
+            if predicted_indices is not None:
+                assert self.n_classes is not None
+                with torch.no_grad():
+                    perplexity, cluster_usage = measure_perplexity(predicted_indices, self.n_classes)
+                log[f"{split}/perplexity"] = perplexity
+                log[f"{split}/cluster_usage"] = cluster_usage
+            return loss, log
+
+        if optimizer_idx == 1:
+            # second pass for discriminator update
+            if cond is None:
+                logits_real = self.discriminator(inputs.contiguous().detach())
+                logits_fake = self.discriminator(reconstructions.contiguous().detach())
+            else:
+                logits_real = self.discriminator(torch.cat((inputs.contiguous().detach(), cond), dim=1))
+                logits_fake = self.discriminator(torch.cat((reconstructions.contiguous().detach(), cond), dim=1))
+
+            disc_factor = adopt_weight(self.disc_factor, global_step, threshold=self.discriminator_iter_start)
+            d_loss = disc_factor * self.disc_loss(logits_real, logits_fake)
+
+            log = {"{}/disc_loss".format(split): d_loss.clone().detach().mean(),
+                   "{}/logits_real".format(split): logits_real.detach().mean(),
+                   "{}/logits_fake".format(split): logits_fake.detach().mean()
+                   }
+            return d_loss, log

examples/tutorial/stable_diffusion/ldm/modules/x_transformer.py

+"""shout-out to https://github.com/lucidrains/x-transformers/tree/main/x_transformers"""
+import torch
+from torch import nn, einsum
+import torch.nn.functional as F
+from functools import partial
+from inspect import isfunction
+from collections import namedtuple
+from einops import rearrange, repeat, reduce
+
+# constants
+
+DEFAULT_DIM_HEAD = 64
+
+Intermediates = namedtuple('Intermediates', [
+    'pre_softmax_attn',
+    'post_softmax_attn'
+])
+
+LayerIntermediates = namedtuple('Intermediates', [
+    'hiddens',
+    'attn_intermediates'
+])
+
+
+class AbsolutePositionalEmbedding(nn.Module):
+    def __init__(self, dim, max_seq_len):
+        super().__init__()
+        self.emb = nn.Embedding(max_seq_len, dim)
+        self.init_()
+
+    def init_(self):
+        nn.init.normal_(self.emb.weight, std=0.02)
+
+    def forward(self, x):
+        n = torch.arange(x.shape[1], device=x.device)
+        return self.emb(n)[None, :, :]
+
+
+class FixedPositionalEmbedding(nn.Module):
+    def __init__(self, dim):
+        super().__init__()
+        inv_freq = 1. / (10000 ** (torch.arange(0, dim, 2).float() / dim))
+        self.register_buffer('inv_freq', inv_freq)
+
+    def forward(self, x, seq_dim=1, offset=0):
+        t = torch.arange(x.shape[seq_dim], device=x.device).type_as(self.inv_freq) + offset
+        sinusoid_inp = torch.einsum('i , j -> i j', t, self.inv_freq)
+        emb = torch.cat((sinusoid_inp.sin(), sinusoid_inp.cos()), dim=-1)
+        return emb[None, :, :]
+
+
+# helpers
+
+def exists(val):
+    return val is not None
+
+
+def default(val, d):
+    if exists(val):
+        return val
+    return d() if isfunction(d) else d
+
+
+def always(val):
+    def inner(*args, **kwargs):
+        return val
+    return inner
+
+
+def not_equals(val):
+    def inner(x):
+        return x != val
+    return inner
+
+
+def equals(val):
+    def inner(x):
+        return x == val
+    return inner
+
+
+def max_neg_value(tensor):
+    return -torch.finfo(tensor.dtype).max
+
+
+# keyword argument helpers
+
+def pick_and_pop(keys, d):
+    values = list(map(lambda key: d.pop(key), keys))
+    return dict(zip(keys, values))
+
+
+def group_dict_by_key(cond, d):
+    return_val = [dict(), dict()]
+    for key in d.keys():
+        match = bool(cond(key))
+        ind = int(not match)
+        return_val[ind][key] = d[key]
+    return (*return_val,)
+
+
+def string_begins_with(prefix, str):
+    return str.startswith(prefix)
+
+
+def group_by_key_prefix(prefix, d):
+    return group_dict_by_key(partial(string_begins_with, prefix), d)
+
+
+def groupby_prefix_and_trim(prefix, d):
+    kwargs_with_prefix, kwargs = group_dict_by_key(partial(string_begins_with, prefix), d)
+    kwargs_without_prefix = dict(map(lambda x: (x[0][len(prefix):], x[1]), tuple(kwargs_with_prefix.items())))
+    return kwargs_without_prefix, kwargs
+
+
+# classes
+class Scale(nn.Module):
+    def __init__(self, value, fn):
+        super().__init__()
+        self.value = value
+        self.fn = fn
+
+    def forward(self, x, **kwargs):
+        x, *rest = self.fn(x, **kwargs)
+        return (x * self.value, *rest)
+
+
+class Rezero(nn.Module):
+    def __init__(self, fn):
+        super().__init__()
+        self.fn = fn
+        self.g = nn.Parameter(torch.zeros(1))
+
+    def forward(self, x, **kwargs):
+        x, *rest = self.fn(x, **kwargs)
+        return (x * self.g, *rest)
+
+
+class ScaleNorm(nn.Module):
+    def __init__(self, dim, eps=1e-5):
+        super().__init__()
+        self.scale = dim ** -0.5
+        self.eps = eps
+        self.g = nn.Parameter(torch.ones(1))
+
+    def forward(self, x):
+        norm = torch.norm(x, dim=-1, keepdim=True) * self.scale
+        return x / norm.clamp(min=self.eps) * self.g
+
+
+class RMSNorm(nn.Module):
+    def __init__(self, dim, eps=1e-8):
+        super().__init__()
+        self.scale = dim ** -0.5
+        self.eps = eps
+        self.g = nn.Parameter(torch.ones(dim))
+
+    def forward(self, x):
+        norm = torch.norm(x, dim=-1, keepdim=True) * self.scale
+        return x / norm.clamp(min=self.eps) * self.g
+
+
+class Residual(nn.Module):
+    def forward(self, x, residual):
+        return x + residual
+
+
+class GRUGating(nn.Module):
+    def __init__(self, dim):
+        super().__init__()
+        self.gru = nn.GRUCell(dim, dim)
+
+    def forward(self, x, residual):
+        gated_output = self.gru(
+            rearrange(x, 'b n d -> (b n) d'),
+            rearrange(residual, 'b n d -> (b n) d')
+        )
+
+        return gated_output.reshape_as(x)
+
+
+# feedforward
+
+class GEGLU(nn.Module):
+    def __init__(self, dim_in, dim_out):
+        super().__init__()
+        self.proj = nn.Linear(dim_in, dim_out * 2)
+
+    def forward(self, x):
+        x, gate = self.proj(x).chunk(2, dim=-1)
+        return x * F.gelu(gate)
+
+
+class FeedForward(nn.Module):
+    def __init__(self, dim, dim_out=None, mult=4, glu=False, dropout=0.):
+        super().__init__()
+        inner_dim = int(dim * mult)
+        dim_out = default(dim_out, dim)
+        project_in = nn.Sequential(
+            nn.Linear(dim, inner_dim),
+            nn.GELU()
+        ) if not glu else GEGLU(dim, inner_dim)
+
+        self.net = nn.Sequential(
+            project_in,
+            nn.Dropout(dropout),
+            nn.Linear(inner_dim, dim_out)
+        )
+
+    def forward(self, x):
+        return self.net(x)
+
+
+# attention.
+class Attention(nn.Module):
+    def __init__(
+            self,
+            dim,
+            dim_head=DEFAULT_DIM_HEAD,
+            heads=8,
+            causal=False,
+            mask=None,
+            talking_heads=False,
+            sparse_topk=None,
+            use_entmax15=False,
+            num_mem_kv=0,
+            dropout=0.,
+            on_attn=False
+    ):
+        super().__init__()
+        if use_entmax15:
+            raise NotImplementedError("Check out entmax activation instead of softmax activation!")
+        self.scale = dim_head ** -0.5
+        self.heads = heads
+        self.causal = causal
+        self.mask = mask
+
+        inner_dim = dim_head * heads
+
+        self.to_q = nn.Linear(dim, inner_dim, bias=False)
+        self.to_k = nn.Linear(dim, inner_dim, bias=False)
+        self.to_v = nn.Linear(dim, inner_dim, bias=False)
+        self.dropout = nn.Dropout(dropout)
+
+        # talking heads
+        self.talking_heads = talking_heads
+        if talking_heads:
+            self.pre_softmax_proj = nn.Parameter(torch.randn(heads, heads))
+            self.post_softmax_proj = nn.Parameter(torch.randn(heads, heads))
+
+        # explicit topk sparse attention
+        self.sparse_topk = sparse_topk
+
+        # entmax
+        #self.attn_fn = entmax15 if use_entmax15 else F.softmax
+        self.attn_fn = F.softmax
+
+        # add memory key / values
+        self.num_mem_kv = num_mem_kv
+        if num_mem_kv > 0:
+            self.mem_k = nn.Parameter(torch.randn(heads, num_mem_kv, dim_head))
+            self.mem_v = nn.Parameter(torch.randn(heads, num_mem_kv, dim_head))
+
+        # attention on attention
+        self.attn_on_attn = on_attn
+        self.to_out = nn.Sequential(nn.Linear(inner_dim, dim * 2), nn.GLU()) if on_attn else nn.Linear(inner_dim, dim)
+
+    def forward(
+            self,
+            x,
+            context=None,
+            mask=None,
+            context_mask=None,
+            rel_pos=None,
+            sinusoidal_emb=None,
+            prev_attn=None,
+            mem=None
+    ):
+        b, n, _, h, talking_heads, device = *x.shape, self.heads, self.talking_heads, x.device
+        kv_input = default(context, x)
+
+        q_input = x
+        k_input = kv_input
+        v_input = kv_input
+
+        if exists(mem):
+            k_input = torch.cat((mem, k_input), dim=-2)
+            v_input = torch.cat((mem, v_input), dim=-2)
+
+        if exists(sinusoidal_emb):
+            # in shortformer, the query would start at a position offset depending on the past cached memory
+            offset = k_input.shape[-2] - q_input.shape[-2]
+            q_input = q_input + sinusoidal_emb(q_input, offset=offset)
+            k_input = k_input + sinusoidal_emb(k_input)
+
+        q = self.to_q(q_input)
+        k = self.to_k(k_input)
+        v = self.to_v(v_input)
+
+        q, k, v = map(lambda t: rearrange(t, 'b n (h d) -> b h n d', h=h), (q, k, v))
+
+        input_mask = None
+        if any(map(exists, (mask, context_mask))):
+            q_mask = default(mask, lambda: torch.ones((b, n), device=device).bool())
+            k_mask = q_mask if not exists(context) else context_mask
+            k_mask = default(k_mask, lambda: torch.ones((b, k.shape[-2]), device=device).bool())
+            q_mask = rearrange(q_mask, 'b i -> b () i ()')
+            k_mask = rearrange(k_mask, 'b j -> b () () j')
+            input_mask = q_mask * k_mask
+
+        if self.num_mem_kv > 0:
+            mem_k, mem_v = map(lambda t: repeat(t, 'h n d -> b h n d', b=b), (self.mem_k, self.mem_v))
+            k = torch.cat((mem_k, k), dim=-2)
+            v = torch.cat((mem_v, v), dim=-2)
+            if exists(input_mask):
+                input_mask = F.pad(input_mask, (self.num_mem_kv, 0), value=True)
+
+        dots = einsum('b h i d, b h j d -> b h i j', q, k) * self.scale
+        mask_value = max_neg_value(dots)
+
+        if exists(prev_attn):
+            dots = dots + prev_attn
+
+        pre_softmax_attn = dots
+
+        if talking_heads:
+            dots = einsum('b h i j, h k -> b k i j', dots, self.pre_softmax_proj).contiguous()
+
+        if exists(rel_pos):
+            dots = rel_pos(dots)
+
+        if exists(input_mask):
+            dots.masked_fill_(~input_mask, mask_value)
+            del input_mask
+
+        if self.causal:
+            i, j = dots.shape[-2:]
+            r = torch.arange(i, device=device)
+            mask = rearrange(r, 'i -> () () i ()') < rearrange(r, 'j -> () () () j')
+            mask = F.pad(mask, (j - i, 0), value=False)
+            dots.masked_fill_(mask, mask_value)
+            del mask
+
+        if exists(self.sparse_topk) and self.sparse_topk < dots.shape[-1]:
+            top, _ = dots.topk(self.sparse_topk, dim=-1)
+            vk = top[..., -1].unsqueeze(-1).expand_as(dots)
+            mask = dots < vk
+            dots.masked_fill_(mask, mask_value)
+            del mask
+
+        attn = self.attn_fn(dots, dim=-1)
+        post_softmax_attn = attn
+
+        attn = self.dropout(attn)
+
+        if talking_heads:
+            attn = einsum('b h i j, h k -> b k i j', attn, self.post_softmax_proj).contiguous()
+
+        out = einsum('b h i j, b h j d -> b h i d', attn, v)
+        out = rearrange(out, 'b h n d -> b n (h d)')
+
+        intermediates = Intermediates(
+            pre_softmax_attn=pre_softmax_attn,
+            post_softmax_attn=post_softmax_attn
+        )
+
+        return self.to_out(out), intermediates
+
+
+class AttentionLayers(nn.Module):
+    def __init__(
+            self,
+            dim,
+            depth,
+            heads=8,
+            causal=False,
+            cross_attend=False,
+            only_cross=False,
+            use_scalenorm=False,
+            use_rmsnorm=False,
+            use_rezero=False,
+            rel_pos_num_buckets=32,
+            rel_pos_max_distance=128,
+            position_infused_attn=False,
+            custom_layers=None,
+            sandwich_coef=None,
+            par_ratio=None,
+            residual_attn=False,
+            cross_residual_attn=False,
+            macaron=False,
+            pre_norm=True,
+            gate_residual=False,
+            **kwargs
+    ):
+        super().__init__()
+        ff_kwargs, kwargs = groupby_prefix_and_trim('ff_', kwargs)
+        attn_kwargs, _ = groupby_prefix_and_trim('attn_', kwargs)
+
+        dim_head = attn_kwargs.get('dim_head', DEFAULT_DIM_HEAD)
+
+        self.dim = dim
+        self.depth = depth
+        self.layers = nn.ModuleList([])
+
+        self.has_pos_emb = position_infused_attn
+        self.pia_pos_emb = FixedPositionalEmbedding(dim) if position_infused_attn else None
+        self.rotary_pos_emb = always(None)
+
+        assert rel_pos_num_buckets <= rel_pos_max_distance, 'number of relative position buckets must be less than the relative position max distance'
+        self.rel_pos = None
+
+        self.pre_norm = pre_norm
+
+        self.residual_attn = residual_attn
+        self.cross_residual_attn = cross_residual_attn
+
+        norm_class = ScaleNorm if use_scalenorm else nn.LayerNorm
+        norm_class = RMSNorm if use_rmsnorm else norm_class
+        norm_fn = partial(norm_class, dim)
+
+        norm_fn = nn.Identity if use_rezero else norm_fn
+        branch_fn = Rezero if use_rezero else None
+
+        if cross_attend and not only_cross:
+            default_block = ('a', 'c', 'f')
+        elif cross_attend and only_cross:
+            default_block = ('c', 'f')
+        else:
+            default_block = ('a', 'f')
+
+        if macaron:
+            default_block = ('f',) + default_block
+
+        if exists(custom_layers):
+            layer_types = custom_layers
+        elif exists(par_ratio):
+            par_depth = depth * len(default_block)
+            assert 1 < par_ratio <= par_depth, 'par ratio out of range'
+            default_block = tuple(filter(not_equals('f'), default_block))
+            par_attn = par_depth // par_ratio
+            depth_cut = par_depth * 2 // 3  # 2 / 3 attention layer cutoff suggested by PAR paper
+            par_width = (depth_cut + depth_cut // par_attn) // par_attn
+            assert len(default_block) <= par_width, 'default block is too large for par_ratio'
+            par_block = default_block + ('f',) * (par_width - len(default_block))
+            par_head = par_block * par_attn
+            layer_types = par_head + ('f',) * (par_depth - len(par_head))
+        elif exists(sandwich_coef):
+            assert sandwich_coef > 0 and sandwich_coef <= depth, 'sandwich coefficient should be less than the depth'
+            layer_types = ('a',) * sandwich_coef + default_block * (depth - sandwich_coef) + ('f',) * sandwich_coef
+        else:
+            layer_types = default_block * depth
+
+        self.layer_types = layer_types
+        self.num_attn_layers = len(list(filter(equals('a'), layer_types)))
+
+        for layer_type in self.layer_types:
+            if layer_type == 'a':
+                layer = Attention(dim, heads=heads, causal=causal, **attn_kwargs)
+            elif layer_type == 'c':
+                layer = Attention(dim, heads=heads, **attn_kwargs)
+            elif layer_type == 'f':
+                layer = FeedForward(dim, **ff_kwargs)
+                layer = layer if not macaron else Scale(0.5, layer)
+            else:
+                raise Exception(f'invalid layer type {layer_type}')
+
+            if isinstance(layer, Attention) and exists(branch_fn):
+                layer = branch_fn(layer)
+
+            if gate_residual:
+                residual_fn = GRUGating(dim)
+            else:
+                residual_fn = Residual()
+
+            self.layers.append(nn.ModuleList([
+                norm_fn(),
+                layer,
+                residual_fn
+            ]))
+
+    def forward(
+            self,
+            x,
+            context=None,
+            mask=None,
+            context_mask=None,
+            mems=None,
+            return_hiddens=False
+    ):
+        hiddens = []
+        intermediates = []
+        prev_attn = None
+        prev_cross_attn = None
+
+        mems = mems.copy() if exists(mems) else [None] * self.num_attn_layers
+
+        for ind, (layer_type, (norm, block, residual_fn)) in enumerate(zip(self.layer_types, self.layers)):
+            is_last = ind == (len(self.layers) - 1)
+
+            if layer_type == 'a':
+                hiddens.append(x)
+                layer_mem = mems.pop(0)
+
+            residual = x
+
+            if self.pre_norm:
+                x = norm(x)
+
+            if layer_type == 'a':
+                out, inter = block(x, mask=mask, sinusoidal_emb=self.pia_pos_emb, rel_pos=self.rel_pos,
+                                   prev_attn=prev_attn, mem=layer_mem)
+            elif layer_type == 'c':
+                out, inter = block(x, context=context, mask=mask, context_mask=context_mask, prev_attn=prev_cross_attn)
+            elif layer_type == 'f':
+                out = block(x)
+
+            x = residual_fn(out, residual)
+
+            if layer_type in ('a', 'c'):
+                intermediates.append(inter)
+
+            if layer_type == 'a' and self.residual_attn:
+                prev_attn = inter.pre_softmax_attn
+            elif layer_type == 'c' and self.cross_residual_attn:
+                prev_cross_attn = inter.pre_softmax_attn
+
+            if not self.pre_norm and not is_last:
+                x = norm(x)
+
+        if return_hiddens:
+            intermediates = LayerIntermediates(
+                hiddens=hiddens,
+                attn_intermediates=intermediates
+            )
+
+            return x, intermediates
+
+        return x
+
+
+class Encoder(AttentionLayers):
+    def __init__(self, **kwargs):
+        assert 'causal' not in kwargs, 'cannot set causality on encoder'
+        super().__init__(causal=False, **kwargs)
+
+
+
+class TransformerWrapper(nn.Module):
+    def __init__(
+            self,
+            *,
+            num_tokens,
+            max_seq_len,
+            attn_layers,
+            emb_dim=None,
+            max_mem_len=0.,
+            emb_dropout=0.,
+            num_memory_tokens=None,
+            tie_embedding=False,
+            use_pos_emb=True
+    ):
+        super().__init__()
+        assert isinstance(attn_layers, AttentionLayers), 'attention layers must be one of Encoder or Decoder'
+
+        dim = attn_layers.dim
+        emb_dim = default(emb_dim, dim)
+
+        self.max_seq_len = max_seq_len
+        self.max_mem_len = max_mem_len
+        self.num_tokens = num_tokens
+
+        self.token_emb = nn.Embedding(num_tokens, emb_dim)
+        self.pos_emb = AbsolutePositionalEmbedding(emb_dim, max_seq_len) if (
+                    use_pos_emb and not attn_layers.has_pos_emb) else always(0)
+        self.emb_dropout = nn.Dropout(emb_dropout)
+
+        self.project_emb = nn.Linear(emb_dim, dim) if emb_dim != dim else nn.Identity()
+        self.attn_layers = attn_layers
+        self.norm = nn.LayerNorm(dim)
+
+        self.init_()
+
+        self.to_logits = nn.Linear(dim, num_tokens) if not tie_embedding else lambda t: t @ self.token_emb.weight.t()
+
+        # memory tokens (like [cls]) from Memory Transformers paper
+        num_memory_tokens = default(num_memory_tokens, 0)
+        self.num_memory_tokens = num_memory_tokens
+        if num_memory_tokens > 0:
+            self.memory_tokens = nn.Parameter(torch.randn(num_memory_tokens, dim))
+
+            # let funnel encoder know number of memory tokens, if specified
+            if hasattr(attn_layers, 'num_memory_tokens'):
+                attn_layers.num_memory_tokens = num_memory_tokens
+
+    def init_(self):
+        nn.init.normal_(self.token_emb.weight, std=0.02)
+
+    def forward(
+            self,
+            x,
+            return_embeddings=False,
+            mask=None,
+            return_mems=False,
+            return_attn=False,
+            mems=None,
+            **kwargs
+    ):
+        b, n, device, num_mem = *x.shape, x.device, self.num_memory_tokens
+        x = self.token_emb(x)
+        x += self.pos_emb(x)
+        x = self.emb_dropout(x)
+
+        x = self.project_emb(x)
+
+        if num_mem > 0:
+            mem = repeat(self.memory_tokens, 'n d -> b n d', b=b)
+            x = torch.cat((mem, x), dim=1)
+
+            # auto-handle masking after appending memory tokens
+            if exists(mask):
+                mask = F.pad(mask, (num_mem, 0), value=True)
+
+        x, intermediates = self.attn_layers(x, mask=mask, mems=mems, return_hiddens=True, **kwargs)
+        x = self.norm(x)
+
+        mem, x = x[:, :num_mem], x[:, num_mem:]
+
+        out = self.to_logits(x) if not return_embeddings else x
+
+        if return_mems:
+            hiddens = intermediates.hiddens
+            new_mems = list(map(lambda pair: torch.cat(pair, dim=-2), zip(mems, hiddens))) if exists(mems) else hiddens
+            new_mems = list(map(lambda t: t[..., -self.max_mem_len:, :].detach(), new_mems))
+            return out, new_mems
+
+        if return_attn:
+            attn_maps = list(map(lambda t: t.post_softmax_attn, intermediates.attn_intermediates))
+            return out, attn_maps
+
+        return out
+

examples/tutorial/stable_diffusion/ldm/util.py

+import importlib
+
+import torch
+import numpy as np
+from collections import abc
+from einops import rearrange
+from functools import partial
+
+import multiprocessing as mp
+from threading import Thread
+from queue import Queue
+
+from inspect import isfunction
+from PIL import Image, ImageDraw, ImageFont
+
+
+def log_txt_as_img(wh, xc, size=10):
+    # wh a tuple of (width, height)
+    # xc a list of captions to plot
+    b = len(xc)
+    txts = list()
+    for bi in range(b):
+        txt = Image.new("RGB", wh, color="white")
+        draw = ImageDraw.Draw(txt)
+        font = ImageFont.truetype('data/DejaVuSans.ttf', size=size)
+        nc = int(40 * (wh[0] / 256))
+        lines = "\n".join(xc[bi][start:start + nc] for start in range(0, len(xc[bi]), nc))
+
+        try:
+            draw.text((0, 0), lines, fill="black", font=font)
+        except UnicodeEncodeError:
+            print("Cant encode string for logging. Skipping.")
+
+        txt = np.array(txt).transpose(2, 0, 1) / 127.5 - 1.0
+        txts.append(txt)
+    txts = np.stack(txts)
+    txts = torch.tensor(txts)
+    return txts
+
+
+def ismap(x):
+    if not isinstance(x, torch.Tensor):
+        return False
+    return (len(x.shape) == 4) and (x.shape[1] > 3)
+
+
+def isimage(x):
+    if not isinstance(x, torch.Tensor):
+        return False
+    return (len(x.shape) == 4) and (x.shape[1] == 3 or x.shape[1] == 1)
+
+
+def exists(x):
+    return x is not None
+
+
+def default(val, d):
+    if exists(val):
+        return val
+    return d() if isfunction(d) else d
+
+
+def mean_flat(tensor):
+    """
+    https://github.com/openai/guided-diffusion/blob/27c20a8fab9cb472df5d6bdd6c8d11c8f430b924/guided_diffusion/nn.py#L86
+    Take the mean over all non-batch dimensions.
+    """
+    return tensor.mean(dim=list(range(1, len(tensor.shape))))
+
+
+def count_params(model, verbose=False):
+    total_params = sum(p.numel() for p in model.parameters())
+    if verbose:
+        print(f"{model.__class__.__name__} has {total_params * 1.e-6:.2f} M params.")
+    return total_params
+
+
+def instantiate_from_config(config):
+    if not "target" in config:
+        if config == '__is_first_stage__':
+            return None
+        elif config == "__is_unconditional__":
+            return None
+        raise KeyError("Expected key `target` to instantiate.")
+    return get_obj_from_str(config["target"])(**config.get("params", dict()))
+
+
+def get_obj_from_str(string, reload=False):
+    module, cls = string.rsplit(".", 1)
+    if reload:
+        module_imp = importlib.import_module(module)
+        importlib.reload(module_imp)
+    return getattr(importlib.import_module(module, package=None), cls)
+
+
+def _do_parallel_data_prefetch(func, Q, data, idx, idx_to_fn=False):
+    # create dummy dataset instance
+
+    # run prefetching
+    if idx_to_fn:
+        res = func(data, worker_id=idx)
+    else:
+        res = func(data)
+    Q.put([idx, res])
+    Q.put("Done")
+
+
+def parallel_data_prefetch(
+        func: callable, data, n_proc, target_data_type="ndarray", cpu_intensive=True, use_worker_id=False
+):
+    # if target_data_type not in ["ndarray", "list"]:
+    #     raise ValueError(
+    #         "Data, which is passed to parallel_data_prefetch has to be either of type list or ndarray."
+    #     )
+    if isinstance(data, np.ndarray) and target_data_type == "list":
+        raise ValueError("list expected but function got ndarray.")
+    elif isinstance(data, abc.Iterable):
+        if isinstance(data, dict):
+            print(
+                f'WARNING:"data" argument passed to parallel_data_prefetch is a dict: Using only its values and disregarding keys.'
+            )
+            data = list(data.values())
+        if target_data_type == "ndarray":
+            data = np.asarray(data)
+        else:
+            data = list(data)
+    else:
+        raise TypeError(
+            f"The data, that shall be processed parallel has to be either an np.ndarray or an Iterable, but is actually {type(data)}."
+        )
+
+    if cpu_intensive:
+        Q = mp.Queue(1000)
+        proc = mp.Process
+    else:
+        Q = Queue(1000)
+        proc = Thread
+    # spawn processes
+    if target_data_type == "ndarray":
+        arguments = [
+            [func, Q, part, i, use_worker_id]
+            for i, part in enumerate(np.array_split(data, n_proc))
+        ]
+    else:
+        step = (
+            int(len(data) / n_proc + 1)
+            if len(data) % n_proc != 0
+            else int(len(data) / n_proc)
+        )
+        arguments = [
+            [func, Q, part, i, use_worker_id]
+            for i, part in enumerate(
+                [data[i: i + step] for i in range(0, len(data), step)]
+            )
+        ]
+    processes = []
+    for i in range(n_proc):
+        p = proc(target=_do_parallel_data_prefetch, args=arguments[i])
+        processes += [p]
+
+    # start processes
+    print(f"Start prefetching...")
+    import time
+
+    start = time.time()
+    gather_res = [[] for _ in range(n_proc)]
+    try:
+        for p in processes:
+            p.start()
+
+        k = 0
+        while k < n_proc:
+            # get result
+            res = Q.get()
+            if res == "Done":
+                k += 1
+            else:
+                gather_res[res[0]] = res[1]
+
+    except Exception as e:
+        print("Exception: ", e)
+        for p in processes:
+            p.terminate()
+
+        raise e
+    finally:
+        for p in processes:
+            p.join()
+        print(f"Prefetching complete. [{time.time() - start} sec.]")
+
+    if target_data_type == 'ndarray':
+        if not isinstance(gather_res[0], np.ndarray):
+            return np.concatenate([np.asarray(r) for r in gather_res], axis=0)
+
+        # order outputs
+        return np.concatenate(gather_res, axis=0)
+    elif target_data_type == 'list':
+        out = []
+        for r in gather_res:
+            out.extend(r)
+        return out
+    else:
+        return gather_res

examples/tutorial/stable_diffusion/main.py

+import argparse, os, sys, datetime, glob, importlib, csv
+import numpy as np
+import time
+import torch
+import torchvision
+import pytorch_lightning as pl
+
+from packaging import version
+from omegaconf import OmegaConf
+from torch.utils.data import random_split, DataLoader, Dataset, Subset
+from functools import partial
+from PIL import Image
+# from pytorch_lightning.strategies.colossalai import ColossalAIStrategy
+# from colossalai.nn.lr_scheduler import CosineAnnealingWarmupLR
+from colossalai.nn.optimizer import HybridAdam
+from prefetch_generator import BackgroundGenerator
+
+from pytorch_lightning import seed_everything
+from pytorch_lightning.trainer import Trainer
+from pytorch_lightning.callbacks import ModelCheckpoint, Callback, LearningRateMonitor
+from pytorch_lightning.utilities.rank_zero import rank_zero_only
+from pytorch_lightning.utilities import rank_zero_info
+from diffusers.models.unet_2d import UNet2DModel
+
+from clip.model import Bottleneck
+from transformers.models.clip.modeling_clip import CLIPTextTransformer
+
+from ldm.data.base import Txt2ImgIterableBaseDataset
+from ldm.util import instantiate_from_config
+import clip
+from einops import rearrange, repeat
+from transformers import CLIPTokenizer, CLIPTextModel
+import kornia
+
+from ldm.modules.x_transformer import *
+from ldm.modules.encoders.modules import *
+from taming.modules.diffusionmodules.model import ResnetBlock
+from taming.modules.transformer.mingpt import *
+from taming.modules.transformer.permuter import *
+
+
+from ldm.modules.ema import LitEma
+from ldm.modules.distributions.distributions import normal_kl, DiagonalGaussianDistribution
+from ldm.models.autoencoder import AutoencoderKL
+from ldm.models.autoencoder import *
+from ldm.models.diffusion.ddim import *
+from ldm.modules.diffusionmodules.openaimodel import *
+from ldm.modules.diffusionmodules.model import *
+from ldm.modules.diffusionmodules.model import Decoder, Encoder, Up_module, Down_module, Mid_module, temb_module
+from ldm.modules.attention import enable_flash_attention
+
+class DataLoaderX(DataLoader):
+
+    def __iter__(self):
+        return BackgroundGenerator(super().__iter__())
+
+
+def get_parser(**parser_kwargs):
+    def str2bool(v):
+        if isinstance(v, bool):
+            return v
+        if v.lower() in ("yes", "true", "t", "y", "1"):
+            return True
+        elif v.lower() in ("no", "false", "f", "n", "0"):
+            return False
+        else:
+            raise argparse.ArgumentTypeError("Boolean value expected.")
+
+    parser = argparse.ArgumentParser(**parser_kwargs)
+    parser.add_argument(
+        "-n",
+        "--name",
+        type=str,
+        const=True,
+        default="",
+        nargs="?",
+        help="postfix for logdir",
+    )
+    parser.add_argument(
+        "-r",
+        "--resume",
+        type=str,
+        const=True,
+        default="",
+        nargs="?",
+        help="resume from logdir or checkpoint in logdir",
+    )
+    parser.add_argument(
+        "-b",
+        "--base",
+        nargs="*",
+        metavar="base_config.yaml",
+        help="paths to base configs. Loaded from left-to-right. "
+             "Parameters can be overwritten or added with command-line options of the form `--key value`.",
+        default=list(),
+    )
+    parser.add_argument(
+        "-t",
+        "--train",
+        type=str2bool,
+        const=True,
+        default=False,
+        nargs="?",
+        help="train",
+    )
+    parser.add_argument(
+        "--no-test",
+        type=str2bool,
+        const=True,
+        default=False,
+        nargs="?",
+        help="disable test",
+    )
+    parser.add_argument(
+        "-p",
+        "--project",
+        help="name of new or path to existing project"
+    )
+    parser.add_argument(
+        "-d",
+        "--debug",
+        type=str2bool,
+        nargs="?",
+        const=True,
+        default=False,
+        help="enable post-mortem debugging",
+    )
+    parser.add_argument(
+        "-s",
+        "--seed",
+        type=int,
+        default=23,
+        help="seed for seed_everything",
+    )
+    parser.add_argument(
+        "-f",
+        "--postfix",
+        type=str,
+        default="",
+        help="post-postfix for default name",
+    )
+    parser.add_argument(
+        "-l",
+        "--logdir",
+        type=str,
+        default="logs",
+        help="directory for logging dat shit",
+    )
+    parser.add_argument(
+        "--scale_lr",
+        type=str2bool,
+        nargs="?",
+        const=True,
+        default=True,
+        help="scale base-lr by ngpu * batch_size * n_accumulate",
+    )
+    parser.add_argument(
+        "--use_fp16",
+        type=str2bool,
+        nargs="?",
+        const=True,
+        default=True,
+        help="whether to use fp16",
+    )
+    parser.add_argument(
+        "--flash",
+        type=str2bool,
+        const=True,
+        default=False,
+        nargs="?",
+        help="whether to use flash attention",
+    )
+    return parser
+
+
+def nondefault_trainer_args(opt):
+    parser = argparse.ArgumentParser()
+    parser = Trainer.add_argparse_args(parser)
+    args = parser.parse_args([])
+    return sorted(k for k in vars(args) if getattr(opt, k) != getattr(args, k))
+
+
+class WrappedDataset(Dataset):
+    """Wraps an arbitrary object with __len__ and __getitem__ into a pytorch dataset"""
+
+    def __init__(self, dataset):
+        self.data = dataset
+
+    def __len__(self):
+        return len(self.data)
+
+    def __getitem__(self, idx):
+        return self.data[idx]
+
+
+def worker_init_fn(_):
+    worker_info = torch.utils.data.get_worker_info()
+
+    dataset = worker_info.dataset
+    worker_id = worker_info.id
+
+    if isinstance(dataset, Txt2ImgIterableBaseDataset):
+        split_size = dataset.num_records // worker_info.num_workers
+        # reset num_records to the true number to retain reliable length information
+        dataset.sample_ids = dataset.valid_ids[worker_id * split_size:(worker_id + 1) * split_size]
+        current_id = np.random.choice(len(np.random.get_state()[1]), 1)
+        return np.random.seed(np.random.get_state()[1][current_id] + worker_id)
+    else:
+        return np.random.seed(np.random.get_state()[1][0] + worker_id)
+
+
+class DataModuleFromConfig(pl.LightningDataModule):
+    def __init__(self, batch_size, train=None, validation=None, test=None, predict=None,
+                 wrap=False, num_workers=None, shuffle_test_loader=False, use_worker_init_fn=False,
+                 shuffle_val_dataloader=False):
+        super().__init__()
+        self.batch_size = batch_size
+        self.dataset_configs = dict()
+        self.num_workers = num_workers if num_workers is not None else batch_size * 2
+        self.use_worker_init_fn = use_worker_init_fn
+        if train is not None:
+            self.dataset_configs["train"] = train
+            self.train_dataloader = self._train_dataloader
+        if validation is not None:
+            self.dataset_configs["validation"] = validation
+            self.val_dataloader = partial(self._val_dataloader, shuffle=shuffle_val_dataloader)
+        if test is not None:
+            self.dataset_configs["test"] = test
+            self.test_dataloader = partial(self._test_dataloader, shuffle=shuffle_test_loader)
+        if predict is not None:
+            self.dataset_configs["predict"] = predict
+            self.predict_dataloader = self._predict_dataloader
+        self.wrap = wrap
+
+    def prepare_data(self):
+        for data_cfg in self.dataset_configs.values():
+            instantiate_from_config(data_cfg)
+
+    def setup(self, stage=None):
+        self.datasets = dict(
+            (k, instantiate_from_config(self.dataset_configs[k]))
+            for k in self.dataset_configs)
+        if self.wrap:
+            for k in self.datasets:
+                self.datasets[k] = WrappedDataset(self.datasets[k])
+
+    def _train_dataloader(self):
+        is_iterable_dataset = isinstance(self.datasets['train'], Txt2ImgIterableBaseDataset)
+        if is_iterable_dataset or self.use_worker_init_fn:
+            init_fn = worker_init_fn
+        else:
+            init_fn = None
+        return DataLoaderX(self.datasets["train"], batch_size=self.batch_size,
+                          num_workers=self.num_workers, shuffle=False if is_iterable_dataset else True,
+                          worker_init_fn=init_fn)
+
+    def _val_dataloader(self, shuffle=False):
+        if isinstance(self.datasets['validation'], Txt2ImgIterableBaseDataset) or self.use_worker_init_fn:
+            init_fn = worker_init_fn
+        else:
+            init_fn = None
+        return DataLoaderX(self.datasets["validation"],
+                          batch_size=self.batch_size,
+                          num_workers=self.num_workers,
+                          worker_init_fn=init_fn,
+                          shuffle=shuffle)
+
+    def _test_dataloader(self, shuffle=False):
+        is_iterable_dataset = isinstance(self.datasets['train'], Txt2ImgIterableBaseDataset)
+        if is_iterable_dataset or self.use_worker_init_fn:
+            init_fn = worker_init_fn
+        else:
+            init_fn = None
+
+        # do not shuffle dataloader for iterable dataset
+        shuffle = shuffle and (not is_iterable_dataset)
+
+        return DataLoaderX(self.datasets["test"], batch_size=self.batch_size,
+                          num_workers=self.num_workers, worker_init_fn=init_fn, shuffle=shuffle)
+
+    def _predict_dataloader(self, shuffle=False):
+        if isinstance(self.datasets['predict'], Txt2ImgIterableBaseDataset) or self.use_worker_init_fn:
+            init_fn = worker_init_fn
+        else:
+            init_fn = None
+        return DataLoaderX(self.datasets["predict"], batch_size=self.batch_size,
+                          num_workers=self.num_workers, worker_init_fn=init_fn)
+
+
+class SetupCallback(Callback):
+    def __init__(self, resume, now, logdir, ckptdir, cfgdir, config, lightning_config):
+        super().__init__()
+        self.resume = resume
+        self.now = now
+        self.logdir = logdir
+        self.ckptdir = ckptdir
+        self.cfgdir = cfgdir
+        self.config = config
+        self.lightning_config = lightning_config
+
+    def on_keyboard_interrupt(self, trainer, pl_module):
+        if trainer.global_rank == 0:
+            print("Summoning checkpoint.")
+            ckpt_path = os.path.join(self.ckptdir, "last.ckpt")
+            trainer.save_checkpoint(ckpt_path)
+
+    # def on_pretrain_routine_start(self, trainer, pl_module):
+    def on_fit_start(self, trainer, pl_module):
+        if trainer.global_rank == 0:
+            # Create logdirs and save configs
+            os.makedirs(self.logdir, exist_ok=True)
+            os.makedirs(self.ckptdir, exist_ok=True)
+            os.makedirs(self.cfgdir, exist_ok=True)
+
+            if "callbacks" in self.lightning_config:
+                if 'metrics_over_trainsteps_checkpoint' in self.lightning_config['callbacks']:
+                    os.makedirs(os.path.join(self.ckptdir, 'trainstep_checkpoints'), exist_ok=True)
+            print("Project config")
+            print(OmegaConf.to_yaml(self.config))
+            OmegaConf.save(self.config,
+                           os.path.join(self.cfgdir, "{}-project.yaml".format(self.now)))
+
+            print("Lightning config")
+            print(OmegaConf.to_yaml(self.lightning_config))
+            OmegaConf.save(OmegaConf.create({"lightning": self.lightning_config}),
+                           os.path.join(self.cfgdir, "{}-lightning.yaml".format(self.now)))
+
+        else:
+            # ModelCheckpoint callback created log directory --- remove it
+            if not self.resume and os.path.exists(self.logdir):
+                dst, name = os.path.split(self.logdir)
+                dst = os.path.join(dst, "child_runs", name)
+                os.makedirs(os.path.split(dst)[0], exist_ok=True)
+                try:
+                    os.rename(self.logdir, dst)
+                except FileNotFoundError:
+                    pass
+
+
+class ImageLogger(Callback):
+    def __init__(self, batch_frequency, max_images, clamp=True, increase_log_steps=True,
+                 rescale=True, disabled=False, log_on_batch_idx=False, log_first_step=False,
+                 log_images_kwargs=None):
+        super().__init__()
+        self.rescale = rescale
+        self.batch_freq = batch_frequency
+        self.max_images = max_images
+        self.logger_log_images = {
+            pl.loggers.CSVLogger: self._testtube,
+        }
+        self.log_steps = [2 ** n for n in range(int(np.log2(self.batch_freq)) + 1)]
+        if not increase_log_steps:
+            self.log_steps = [self.batch_freq]
+        self.clamp = clamp
+        self.disabled = disabled
+        self.log_on_batch_idx = log_on_batch_idx
+        self.log_images_kwargs = log_images_kwargs if log_images_kwargs else {}
+        self.log_first_step = log_first_step
+
+    @rank_zero_only
+    def _testtube(self, pl_module, images, batch_idx, split):
+        for k in images:
+            grid = torchvision.utils.make_grid(images[k])
+            grid = (grid + 1.0) / 2.0  # -1,1 -> 0,1; c,h,w
+
+            tag = f"{split}/{k}"
+            pl_module.logger.experiment.add_image(
+                tag, grid,
+                global_step=pl_module.global_step)
+
+    @rank_zero_only
+    def log_local(self, save_dir, split, images,
+                  global_step, current_epoch, batch_idx):
+        root = os.path.join(save_dir, "images", split)
+        for k in images:
+            grid = torchvision.utils.make_grid(images[k], nrow=4)
+            if self.rescale:
+                grid = (grid + 1.0) / 2.0  # -1,1 -> 0,1; c,h,w
+            grid = grid.transpose(0, 1).transpose(1, 2).squeeze(-1)
+            grid = grid.numpy()
+            grid = (grid * 255).astype(np.uint8)
+            filename = "{}_gs-{:06}_e-{:06}_b-{:06}.png".format(
+                k,
+                global_step,
+                current_epoch,
+                batch_idx)
+            path = os.path.join(root, filename)
+            os.makedirs(os.path.split(path)[0], exist_ok=True)
+            Image.fromarray(grid).save(path)
+
+    def log_img(self, pl_module, batch, batch_idx, split="train"):
+        check_idx = batch_idx if self.log_on_batch_idx else pl_module.global_step
+        if (self.check_frequency(check_idx) and  # batch_idx % self.batch_freq == 0
+                hasattr(pl_module, "log_images") and
+                callable(pl_module.log_images) and
+                self.max_images > 0):
+            logger = type(pl_module.logger)
+
+            is_train = pl_module.training
+            if is_train:
+                pl_module.eval()
+
+            with torch.no_grad():
+                images = pl_module.log_images(batch, split=split, **self.log_images_kwargs)
+
+            for k in images:
+                N = min(images[k].shape[0], self.max_images)
+                images[k] = images[k][:N]
+                if isinstance(images[k], torch.Tensor):
+                    images[k] = images[k].detach().cpu()
+                    if self.clamp:
+                        images[k] = torch.clamp(images[k], -1., 1.)
+
+            self.log_local(pl_module.logger.save_dir, split, images,
+                           pl_module.global_step, pl_module.current_epoch, batch_idx)
+
+            logger_log_images = self.logger_log_images.get(logger, lambda *args, **kwargs: None)
+            logger_log_images(pl_module, images, pl_module.global_step, split)
+
+            if is_train:
+                pl_module.train()
+
+    def check_frequency(self, check_idx):
+        if ((check_idx % self.batch_freq) == 0 or (check_idx in self.log_steps)) and (
+                check_idx > 0 or self.log_first_step):
+            try:
+                self.log_steps.pop(0)
+            except IndexError as e:
+                print(e)
+                pass
+            return True
+        return False
+
+    def on_train_batch_end(self, trainer, pl_module, outputs, batch, batch_idx):
+        # if not self.disabled and (pl_module.global_step > 0 or self.log_first_step):
+        #     self.log_img(pl_module, batch, batch_idx, split="train")
+        pass
+
+    def on_validation_batch_end(self, trainer, pl_module, outputs, batch, batch_idx):
+        if not self.disabled and pl_module.global_step > 0:
+            self.log_img(pl_module, batch, batch_idx, split="val")
+        if hasattr(pl_module, 'calibrate_grad_norm'):
+            if (pl_module.calibrate_grad_norm and batch_idx % 25 == 0) and batch_idx > 0:
+                self.log_gradients(trainer, pl_module, batch_idx=batch_idx)
+
+
+class CUDACallback(Callback):
+    # see https://github.com/SeanNaren/minGPT/blob/master/mingpt/callback.py
+
+    def on_train_start(self, trainer, pl_module):
+        rank_zero_info("Training is starting")
+
+    def on_train_end(self, trainer, pl_module):
+        rank_zero_info("Training is ending")
+
+    def on_train_epoch_start(self, trainer, pl_module):
+        # Reset the memory use counter
+        torch.cuda.reset_peak_memory_stats(trainer.strategy.root_device.index)
+        torch.cuda.synchronize(trainer.strategy.root_device.index)
+        self.start_time = time.time()
+
+    def on_train_epoch_end(self, trainer, pl_module):
+        torch.cuda.synchronize(trainer.strategy.root_device.index)
+        max_memory = torch.cuda.max_memory_allocated(trainer.strategy.root_device.index) / 2 ** 20
+        epoch_time = time.time() - self.start_time
+
+        try:
+            max_memory = trainer.strategy.reduce(max_memory)
+            epoch_time = trainer.strategy.reduce(epoch_time)
+
+            rank_zero_info(f"Average Epoch time: {epoch_time:.2f} seconds")
+            rank_zero_info(f"Average Peak memory {max_memory:.2f}MiB")
+        except AttributeError:
+            pass
+
+
+if __name__ == "__main__":
+    # custom parser to specify config files, train, test and debug mode,
+    # postfix, resume.
+    # `--key value` arguments are interpreted as arguments to the trainer.
+    # `nested.key=value` arguments are interpreted as config parameters.
+    # configs are merged from left-to-right followed by command line parameters.
+
+    # model:
+    #   base_learning_rate: float
+    #   target: path to lightning module
+    #   params:
+    #       key: value
+    # data:
+    #   target: main.DataModuleFromConfig
+    #   params:
+    #      batch_size: int
+    #      wrap: bool
+    #      train:
+    #          target: path to train dataset
+    #          params:
+    #              key: value
+    #      validation:
+    #          target: path to validation dataset
+    #          params:
+    #              key: value
+    #      test:
+    #          target: path to test dataset
+    #          params:
+    #              key: value
+    # lightning: (optional, has sane defaults and can be specified on cmdline)
+    #   trainer:
+    #       additional arguments to trainer
+    #   logger:
+    #       logger to instantiate
+    #   modelcheckpoint:
+    #       modelcheckpoint to instantiate
+    #   callbacks:
+    #       callback1:
+    #           target: importpath
+    #           params:
+    #               key: value
+
+    now = datetime.datetime.now().strftime("%Y-%m-%dT%H-%M-%S")
+
+    # add cwd for convenience and to make classes in this file available when
+    # running as `python main.py`
+    # (in particular `main.DataModuleFromConfig`)
+    sys.path.append(os.getcwd())
+
+    parser = get_parser()
+    parser = Trainer.add_argparse_args(parser)
+
+    opt, unknown = parser.parse_known_args()
+    if opt.name and opt.resume:
+        raise ValueError(
+            "-n/--name and -r/--resume cannot be specified both."
+            "If you want to resume training in a new log folder, "
+            "use -n/--name in combination with --resume_from_checkpoint"
+        )
+    if opt.flash:
+        enable_flash_attention()
+    if opt.resume:
+        if not os.path.exists(opt.resume):
+            raise ValueError("Cannot find {}".format(opt.resume))
+        if os.path.isfile(opt.resume):
+            paths = opt.resume.split("/")
+            # idx = len(paths)-paths[::-1].index("logs")+1
+            # logdir = "/".join(paths[:idx])
+            logdir = "/".join(paths[:-2])
+            ckpt = opt.resume
+        else:
+            assert os.path.isdir(opt.resume), opt.resume
+            logdir = opt.resume.rstrip("/")
+            ckpt = os.path.join(logdir, "checkpoints", "last.ckpt")
+
+        opt.resume_from_checkpoint = ckpt
+        base_configs = sorted(glob.glob(os.path.join(logdir, "configs/*.yaml")))
+        opt.base = base_configs + opt.base
+        _tmp = logdir.split("/")
+        nowname = _tmp[-1]
+    else:
+        if opt.name:
+            name = "_" + opt.name
+        elif opt.base:
+            cfg_fname = os.path.split(opt.base[0])[-1]
+            cfg_name = os.path.splitext(cfg_fname)[0]
+            name = "_" + cfg_name
+        else:
+            name = ""
+        nowname = now + name + opt.postfix
+        logdir = os.path.join(opt.logdir, nowname)
+
+    ckptdir = os.path.join(logdir, "checkpoints")
+    cfgdir = os.path.join(logdir, "configs")
+    seed_everything(opt.seed)
+
+    try:
+        # init and save configs
+        configs = [OmegaConf.load(cfg) for cfg in opt.base]
+        cli = OmegaConf.from_dotlist(unknown)
+        config = OmegaConf.merge(*configs, cli)
+        lightning_config = config.pop("lightning", OmegaConf.create())
+        # merge trainer cli with config
+        trainer_config = lightning_config.get("trainer", OmegaConf.create())
+  
+        for k in nondefault_trainer_args(opt):
+            trainer_config[k] = getattr(opt, k)
+
+        print(trainer_config)
+        if not trainer_config["accelerator"] == "gpu":
+            del trainer_config["accelerator"]
+            cpu = True
+            print("Running on CPU")
+        else:
+            cpu = False
+            print("Running on GPU")
+        trainer_opt = argparse.Namespace(**trainer_config)
+        lightning_config.trainer = trainer_config
+
+        # model
+        use_fp16 = trainer_config.get("precision", 32) == 16
+        if use_fp16:
+            config.model["params"].update({"use_fp16": True})
+            print("Using FP16 = {}".format(config.model["params"]["use_fp16"]))
+        else:
+            config.model["params"].update({"use_fp16": False})
+            print("Using FP16 = {}".format(config.model["params"]["use_fp16"]))
+        
+        model = instantiate_from_config(config.model)
+        # trainer and callbacks
+        trainer_kwargs = dict()
+
+        # config the logger
+        # default logger configs
+        default_logger_cfgs = {
+            "wandb": {
+                "target": "pytorch_lightning.loggers.WandbLogger",
+                "params": {
+                    "name": nowname,
+                    "save_dir": logdir,
+                    "offline": opt.debug,
+                    "id": nowname,
+                }
+            },
+            "tensorboard":{
+                "target": "pytorch_lightning.loggers.TensorBoardLogger",
+                "params":{
+                    "save_dir": logdir,
+                    "name": "diff_tb",
+                    "log_graph": True
+                }
+            }
+        }
+
+        default_logger_cfg = default_logger_cfgs["tensorboard"]
+        if "logger" in lightning_config:
+            logger_cfg = lightning_config.logger
+        else:
+            logger_cfg = default_logger_cfg
+        logger_cfg = OmegaConf.merge(default_logger_cfg, logger_cfg)
+        trainer_kwargs["logger"] = instantiate_from_config(logger_cfg)
+
+        # config the strategy, defualt is ddp
+        if "strategy" in trainer_config:
+            strategy_cfg = trainer_config["strategy"]
+            print("Using strategy: {}".format(strategy_cfg["target"]))
+        else:
+            strategy_cfg = {
+                "target": "pytorch_lightning.strategies.DDPStrategy",
+                "params": {
+                    "find_unused_parameters": False
+                }
+            }
+            print("Using strategy: DDPStrategy")
+
+        trainer_kwargs["strategy"] = instantiate_from_config(strategy_cfg)
+
+        # modelcheckpoint - use TrainResult/EvalResult(checkpoint_on=metric) to
+        # specify which metric is used to determine best models
+        default_modelckpt_cfg = {
+            "target": "pytorch_lightning.callbacks.ModelCheckpoint",
+            "params": {
+                "dirpath": ckptdir,
+                "filename": "{epoch:06}",
+                "verbose": True,
+                "save_last": True,
+            }
+        }
+        if hasattr(model, "monitor"):
+            print(f"Monitoring {model.monitor} as checkpoint metric.")
+            default_modelckpt_cfg["params"]["monitor"] = model.monitor
+            default_modelckpt_cfg["params"]["save_top_k"] = 3
+
+        if "modelcheckpoint" in lightning_config:
+            modelckpt_cfg = lightning_config.modelcheckpoint
+        else:
+            modelckpt_cfg =  OmegaConf.create()
+        modelckpt_cfg = OmegaConf.merge(default_modelckpt_cfg, modelckpt_cfg)
+        print(f"Merged modelckpt-cfg: \n{modelckpt_cfg}")
+        if version.parse(pl.__version__) < version.parse('1.4.0'):
+            trainer_kwargs["checkpoint_callback"] = instantiate_from_config(modelckpt_cfg)
+
+        # add callback which sets up log directory
+        default_callbacks_cfg = {
+            "setup_callback": {
+                "target": "main.SetupCallback",
+                "params": {
+                    "resume": opt.resume,
+                    "now": now,
+                    "logdir": logdir,
+                    "ckptdir": ckptdir,
+                    "cfgdir": cfgdir,
+                    "config": config,
+                    "lightning_config": lightning_config,
+                }
+            },
+            "image_logger": {
+                "target": "main.ImageLogger",
+                "params": {
+                    "batch_frequency": 750,
+                    "max_images": 4,
+                    "clamp": True
+                }
+            },
+            "learning_rate_logger": {
+                "target": "main.LearningRateMonitor",
+                "params": {
+                    "logging_interval": "step",
+                    # "log_momentum": True
+                }
+            },
+            "cuda_callback": {
+                "target": "main.CUDACallback"
+            },
+        }
+        if version.parse(pl.__version__) >= version.parse('1.4.0'):
+            default_callbacks_cfg.update({'checkpoint_callback': modelckpt_cfg})
+
+        if "callbacks" in lightning_config:
+            callbacks_cfg = lightning_config.callbacks
+        else:
+            callbacks_cfg = OmegaConf.create()
+
+        if 'metrics_over_trainsteps_checkpoint' in callbacks_cfg:
+            print(
+                'Caution: Saving checkpoints every n train steps without deleting. This might require some free space.')
+            default_metrics_over_trainsteps_ckpt_dict = {
+                'metrics_over_trainsteps_checkpoint':
+                    {"target": 'pytorch_lightning.callbacks.ModelCheckpoint',
+                     'params': {
+                         "dirpath": os.path.join(ckptdir, 'trainstep_checkpoints'),
+                         "filename": "{epoch:06}-{step:09}",
+                         "verbose": True,
+                         'save_top_k': -1,
+                         'every_n_train_steps': 10000,
+                         'save_weights_only': True
+                     }
+                     }
+            }
+            default_callbacks_cfg.update(default_metrics_over_trainsteps_ckpt_dict)
+
+        callbacks_cfg = OmegaConf.merge(default_callbacks_cfg, callbacks_cfg)
+        if 'ignore_keys_callback' in callbacks_cfg and hasattr(trainer_opt, 'resume_from_checkpoint'):
+            callbacks_cfg.ignore_keys_callback.params['ckpt_path'] = trainer_opt.resume_from_checkpoint
+        elif 'ignore_keys_callback' in callbacks_cfg:
+            del callbacks_cfg['ignore_keys_callback']
+
+        trainer_kwargs["callbacks"] = [instantiate_from_config(callbacks_cfg[k]) for k in callbacks_cfg]
+
+        trainer = Trainer.from_argparse_args(trainer_opt, **trainer_kwargs)
+        trainer.logdir = logdir  ###
+
+        # data
+        data = instantiate_from_config(config.data)
+        # NOTE according to https://pytorch-lightning.readthedocs.io/en/latest/datamodules.html
+        # calling these ourselves should not be necessary but it is.
+        # lightning still takes care of proper multiprocessing though
+        data.prepare_data()
+        data.setup()
+        print("#### Data #####")
+        for k in data.datasets:
+            print(f"{k}, {data.datasets[k].__class__.__name__}, {len(data.datasets[k])}")
+
+        # configure learning rate
+        bs, base_lr = config.data.params.batch_size, config.model.base_learning_rate
+        if not cpu:
+            ngpu = trainer_config["devices"]
+        else:
+            ngpu = 1
+        if 'accumulate_grad_batches' in lightning_config.trainer:
+            accumulate_grad_batches = lightning_config.trainer.accumulate_grad_batches
+        else:
+            accumulate_grad_batches = 1
+        print(f"accumulate_grad_batches = {accumulate_grad_batches}")
+        lightning_config.trainer.accumulate_grad_batches = accumulate_grad_batches
+        if opt.scale_lr:
+            model.learning_rate = accumulate_grad_batches * ngpu * bs * base_lr
+            print(
+                "Setting learning rate to {:.2e} = {} (accumulate_grad_batches) * {} (num_gpus) * {} (batchsize) * {:.2e} (base_lr)".format(
+                    model.learning_rate, accumulate_grad_batches, ngpu, bs, base_lr))
+        else:
+            model.learning_rate = base_lr
+            print("++++ NOT USING LR SCALING ++++")
+            print(f"Setting learning rate to {model.learning_rate:.2e}")
+
+
+        # allow checkpointing via USR1
+        def melk(*args, **kwargs):
+            # run all checkpoint hooks
+            if trainer.global_rank == 0:
+                print("Summoning checkpoint.")
+                ckpt_path = os.path.join(ckptdir, "last.ckpt")
+                trainer.save_checkpoint(ckpt_path)
+
+
+        def divein(*args, **kwargs):
+            if trainer.global_rank == 0:
+                import pudb;
+                pudb.set_trace()
+
+
+        import signal
+
+        signal.signal(signal.SIGUSR1, melk)
+        signal.signal(signal.SIGUSR2, divein)
+
+        # run
+        if opt.train:
+            try:
+                for name, m in model.named_parameters():
+                    print(name)
+                trainer.fit(model, data)
+            except Exception:
+                melk()
+                raise
+        # if not opt.no_test and not trainer.interrupted:
+        #     trainer.test(model, data)
+    except Exception:
+        if opt.debug and trainer.global_rank == 0:
+            try:
+                import pudb as debugger
+            except ImportError:
+                import pdb as debugger
+            debugger.post_mortem()
+        raise
+    finally:
+        # move newly created debug project to debug_runs
+        if opt.debug and not opt.resume and trainer.global_rank == 0:
+            dst, name = os.path.split(logdir)
+            dst = os.path.join(dst, "debug_runs", name)
+            os.makedirs(os.path.split(dst)[0], exist_ok=True)
+            os.rename(logdir, dst)
+        if trainer.global_rank == 0:
+            print(trainer.profiler.summary())

examples/tutorial/stable_diffusion/requirements.txt

+albumentations==0.4.3
+diffusers
+opencv-python==4.1.2.30
+pudb==2019.2
+invisible-watermark
+imageio==2.9.0
+imageio-ffmpeg==0.4.2
+omegaconf==2.1.1
+test-tube>=0.7.5
+streamlit>=0.73.1
+einops==0.3.0
+torch-fidelity==0.3.0
+transformers==4.19.2
+torchmetrics==0.6.0
+kornia==0.6
+opencv-python==4.6.0.66
+prefetch_generator
+-e git+https://github.com/CompVis/taming-transformers.git@master#egg=taming-transformers
+-e git+https://github.com/openai/CLIP.git@main#egg=clip
+-e .

examples/tutorial/stable_diffusion/scripts/download_first_stages.sh

+#!/bin/bash
+wget -O models/first_stage_models/kl-f4/model.zip https://ommer-lab.com/files/latent-diffusion/kl-f4.zip
+wget -O models/first_stage_models/kl-f8/model.zip https://ommer-lab.com/files/latent-diffusion/kl-f8.zip
+wget -O models/first_stage_models/kl-f16/model.zip https://ommer-lab.com/files/latent-diffusion/kl-f16.zip
+wget -O models/first_stage_models/kl-f32/model.zip https://ommer-lab.com/files/latent-diffusion/kl-f32.zip
+wget -O models/first_stage_models/vq-f4/model.zip https://ommer-lab.com/files/latent-diffusion/vq-f4.zip
+wget -O models/first_stage_models/vq-f4-noattn/model.zip https://ommer-lab.com/files/latent-diffusion/vq-f4-noattn.zip
+wget -O models/first_stage_models/vq-f8/model.zip https://ommer-lab.com/files/latent-diffusion/vq-f8.zip
+wget -O models/first_stage_models/vq-f8-n256/model.zip https://ommer-lab.com/files/latent-diffusion/vq-f8-n256.zip
+wget -O models/first_stage_models/vq-f16/model.zip https://ommer-lab.com/files/latent-diffusion/vq-f16.zip
+
+
+
+cd models/first_stage_models/kl-f4
+unzip -o model.zip
+
+cd ../kl-f8
+unzip -o model.zip
+
+cd ../kl-f16
+unzip -o model.zip
+
+cd ../kl-f32
+unzip -o model.zip
+
+cd ../vq-f4
+unzip -o model.zip
+
+cd ../vq-f4-noattn
+unzip -o model.zip
+
+cd ../vq-f8
+unzip -o model.zip
+
+cd ../vq-f8-n256
+unzip -o model.zip
+
+cd ../vq-f16
+unzip -o model.zip
+
+cd ../..
\ No newline at end of file

examples/tutorial/stable_diffusion/scripts/download_models.sh

+#!/bin/bash
+wget -O models/ldm/celeba256/celeba-256.zip https://ommer-lab.com/files/latent-diffusion/celeba.zip
+wget -O models/ldm/ffhq256/ffhq-256.zip https://ommer-lab.com/files/latent-diffusion/ffhq.zip
+wget -O models/ldm/lsun_churches256/lsun_churches-256.zip https://ommer-lab.com/files/latent-diffusion/lsun_churches.zip
+wget -O models/ldm/lsun_beds256/lsun_beds-256.zip https://ommer-lab.com/files/latent-diffusion/lsun_bedrooms.zip
+wget -O models/ldm/text2img256/model.zip https://ommer-lab.com/files/latent-diffusion/text2img.zip
+wget -O models/ldm/cin256/model.zip https://ommer-lab.com/files/latent-diffusion/cin.zip
+wget -O models/ldm/semantic_synthesis512/model.zip https://ommer-lab.com/files/latent-diffusion/semantic_synthesis.zip
+wget -O models/ldm/semantic_synthesis256/model.zip https://ommer-lab.com/files/latent-diffusion/semantic_synthesis256.zip
+wget -O models/ldm/bsr_sr/model.zip https://ommer-lab.com/files/latent-diffusion/sr_bsr.zip
+wget -O models/ldm/layout2img-openimages256/model.zip https://ommer-lab.com/files/latent-diffusion/layout2img_model.zip
+wget -O models/ldm/inpainting_big/model.zip https://ommer-lab.com/files/latent-diffusion/inpainting_big.zip
+
+
+
+cd models/ldm/celeba256
+unzip -o celeba-256.zip
+
+cd ../ffhq256
+unzip -o ffhq-256.zip
+
+cd ../lsun_churches256
+unzip -o lsun_churches-256.zip
+
+cd ../lsun_beds256
+unzip -o lsun_beds-256.zip
+
+cd ../text2img256
+unzip -o model.zip
+
+cd ../cin256
+unzip -o model.zip
+
+cd ../semantic_synthesis512
+unzip -o model.zip
+
+cd ../semantic_synthesis256
+unzip -o model.zip
+
+cd ../bsr_sr
+unzip -o model.zip
+
+cd ../layout2img-openimages256
+unzip -o model.zip
+
+cd ../inpainting_big
+unzip -o model.zip
+
+cd ../..

examples/tutorial/stable_diffusion/scripts/img2img.py

+"""make variations of input image"""
+
+import argparse, os, sys, glob
+import PIL
+import torch
+import numpy as np
+from omegaconf import OmegaConf
+from PIL import Image
+from tqdm import tqdm, trange
+from itertools import islice
+from einops import rearrange, repeat
+from torchvision.utils import make_grid
+from torch import autocast
+from contextlib import nullcontext
+import time
+from pytorch_lightning import seed_everything
+
+from ldm.util import instantiate_from_config
+from ldm.models.diffusion.ddim import DDIMSampler
+from ldm.models.diffusion.plms import PLMSSampler
+
+
+def chunk(it, size):
+    it = iter(it)
+    return iter(lambda: tuple(islice(it, size)), ())
+
+
+def load_model_from_config(config, ckpt, verbose=False):
+    print(f"Loading model from {ckpt}")
+    pl_sd = torch.load(ckpt, map_location="cpu")
+    if "global_step" in pl_sd:
+        print(f"Global Step: {pl_sd['global_step']}")
+    sd = pl_sd["state_dict"]
+    model = instantiate_from_config(config.model)
+    m, u = model.load_state_dict(sd, strict=False)
+    if len(m) > 0 and verbose:
+        print("missing keys:")
+        print(m)
+    if len(u) > 0 and verbose:
+        print("unexpected keys:")
+        print(u)
+
+    model.cuda()
+    model.eval()
+    return model
+
+
+def load_img(path):
+    image = Image.open(path).convert("RGB")
+    w, h = image.size
+    print(f"loaded input image of size ({w}, {h}) from {path}")
+    w, h = map(lambda x: x - x % 32, (w, h))  # resize to integer multiple of 32
+    image = image.resize((w, h), resample=PIL.Image.LANCZOS)
+    image = np.array(image).astype(np.float32) / 255.0
+    image = image[None].transpose(0, 3, 1, 2)
+    image = torch.from_numpy(image)
+    return 2.*image - 1.
+
+
+def main():
+    parser = argparse.ArgumentParser()
+
+    parser.add_argument(
+        "--prompt",
+        type=str,
+        nargs="?",
+        default="a painting of a virus monster playing guitar",
+        help="the prompt to render"
+    )
+
+    parser.add_argument(
+        "--init-img",
+        type=str,
+        nargs="?",
+        help="path to the input image"
+    )
+
+    parser.add_argument(
+        "--outdir",
+        type=str,
+        nargs="?",
+        help="dir to write results to",
+        default="outputs/img2img-samples"
+    )
+
+    parser.add_argument(
+        "--skip_grid",
+        action='store_true',
+        help="do not save a grid, only individual samples. Helpful when evaluating lots of samples",
+    )
+
+    parser.add_argument(
+        "--skip_save",
+        action='store_true',
+        help="do not save indiviual samples. For speed measurements.",
+    )
+
+    parser.add_argument(
+        "--ddim_steps",
+        type=int,
+        default=50,
+        help="number of ddim sampling steps",
+    )
+
+    parser.add_argument(
+        "--plms",
+        action='store_true',
+        help="use plms sampling",
+    )
+    parser.add_argument(
+        "--fixed_code",
+        action='store_true',
+        help="if enabled, uses the same starting code across all samples ",
+    )
+
+    parser.add_argument(
+        "--ddim_eta",
+        type=float,
+        default=0.0,
+        help="ddim eta (eta=0.0 corresponds to deterministic sampling",
+    )
+    parser.add_argument(
+        "--n_iter",
+        type=int,
+        default=1,
+        help="sample this often",
+    )
+    parser.add_argument(
+        "--C",
+        type=int,
+        default=4,
+        help="latent channels",
+    )
+    parser.add_argument(
+        "--f",
+        type=int,
+        default=8,
+        help="downsampling factor, most often 8 or 16",
+    )
+    parser.add_argument(
+        "--n_samples",
+        type=int,
+        default=2,
+        help="how many samples to produce for each given prompt. A.k.a batch size",
+    )
+    parser.add_argument(
+        "--n_rows",
+        type=int,
+        default=0,
+        help="rows in the grid (default: n_samples)",
+    )
+    parser.add_argument(
+        "--scale",
+        type=float,
+        default=5.0,
+        help="unconditional guidance scale: eps = eps(x, empty) + scale * (eps(x, cond) - eps(x, empty))",
+    )
+
+    parser.add_argument(
+        "--strength",
+        type=float,
+        default=0.75,
+        help="strength for noising/unnoising. 1.0 corresponds to full destruction of information in init image",
+    )
+    parser.add_argument(
+        "--from-file",
+        type=str,
+        help="if specified, load prompts from this file",
+    )
+    parser.add_argument(
+        "--config",
+        type=str,
+        default="configs/stable-diffusion/v1-inference.yaml",
+        help="path to config which constructs model",
+    )
+    parser.add_argument(
+        "--ckpt",
+        type=str,
+        default="models/ldm/stable-diffusion-v1/model.ckpt",
+        help="path to checkpoint of model",
+    )
+    parser.add_argument(
+        "--seed",
+        type=int,
+        default=42,
+        help="the seed (for reproducible sampling)",
+    )
+    parser.add_argument(
+        "--precision",
+        type=str,
+        help="evaluate at this precision",
+        choices=["full", "autocast"],
+        default="autocast"
+    )
+
+    opt = parser.parse_args()
+    seed_everything(opt.seed)
+
+    config = OmegaConf.load(f"{opt.config}")
+    model = load_model_from_config(config, f"{opt.ckpt}")
+
+    device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")
+    model = model.to(device)
+
+    if opt.plms:
+        raise NotImplementedError("PLMS sampler not (yet) supported")
+        sampler = PLMSSampler(model)
+    else:
+        sampler = DDIMSampler(model)
+
+    os.makedirs(opt.outdir, exist_ok=True)
+    outpath = opt.outdir
+
+    batch_size = opt.n_samples
+    n_rows = opt.n_rows if opt.n_rows > 0 else batch_size
+    if not opt.from_file:
+        prompt = opt.prompt
+        assert prompt is not None
+        data = [batch_size * [prompt]]
+
+    else:
+        print(f"reading prompts from {opt.from_file}")
+        with open(opt.from_file, "r") as f:
+            data = f.read().splitlines()
+            data = list(chunk(data, batch_size))
+
+    sample_path = os.path.join(outpath, "samples")
+    os.makedirs(sample_path, exist_ok=True)
+    base_count = len(os.listdir(sample_path))
+    grid_count = len(os.listdir(outpath)) - 1
+
+    assert os.path.isfile(opt.init_img)
+    init_image = load_img(opt.init_img).to(device)
+    init_image = repeat(init_image, '1 ... -> b ...', b=batch_size)
+    init_latent = model.get_first_stage_encoding(model.encode_first_stage(init_image))  # move to latent space
+
+    sampler.make_schedule(ddim_num_steps=opt.ddim_steps, ddim_eta=opt.ddim_eta, verbose=False)
+
+    assert 0. <= opt.strength <= 1., 'can only work with strength in [0.0, 1.0]'
+    t_enc = int(opt.strength * opt.ddim_steps)
+    print(f"target t_enc is {t_enc} steps")
+
+    precision_scope = autocast if opt.precision == "autocast" else nullcontext
+    with torch.no_grad():
+        with precision_scope("cuda"):
+            with model.ema_scope():
+                tic = time.time()
+                all_samples = list()
+                for n in trange(opt.n_iter, desc="Sampling"):
+                    for prompts in tqdm(data, desc="data"):
+                        uc = None
+                        if opt.scale != 1.0:
+                            uc = model.get_learned_conditioning(batch_size * [""])
+                        if isinstance(prompts, tuple):
+                            prompts = list(prompts)
+                        c = model.get_learned_conditioning(prompts)
+
+                        # encode (scaled latent)
+                        z_enc = sampler.stochastic_encode(init_latent, torch.tensor([t_enc]*batch_size).to(device))
+                        # decode it
+                        samples = sampler.decode(z_enc, c, t_enc, unconditional_guidance_scale=opt.scale,
+                                                 unconditional_conditioning=uc,)
+
+                        x_samples = model.decode_first_stage(samples)
+                        x_samples = torch.clamp((x_samples + 1.0) / 2.0, min=0.0, max=1.0)
+
+                        if not opt.skip_save:
+                            for x_sample in x_samples:
+                                x_sample = 255. * rearrange(x_sample.cpu().numpy(), 'c h w -> h w c')
+                                Image.fromarray(x_sample.astype(np.uint8)).save(
+                                    os.path.join(sample_path, f"{base_count:05}.png"))
+                                base_count += 1
+                        all_samples.append(x_samples)
+
+                if not opt.skip_grid:
+                    # additionally, save as grid
+                    grid = torch.stack(all_samples, 0)
+                    grid = rearrange(grid, 'n b c h w -> (n b) c h w')
+                    grid = make_grid(grid, nrow=n_rows)
+
+                    # to image
+                    grid = 255. * rearrange(grid, 'c h w -> h w c').cpu().numpy()
+                    Image.fromarray(grid.astype(np.uint8)).save(os.path.join(outpath, f'grid-{grid_count:04}.png'))
+                    grid_count += 1
+
+                toc = time.time()
+
+    print(f"Your samples are ready and waiting for you here: \n{outpath} \n"
+          f" \nEnjoy.")
+
+
+if __name__ == "__main__":
+    main()

examples/tutorial/stable_diffusion/scripts/inpaint.py

+import argparse, os, sys, glob
+from omegaconf import OmegaConf
+from PIL import Image
+from tqdm import tqdm
+import numpy as np
+import torch
+from main import instantiate_from_config
+from ldm.models.diffusion.ddim import DDIMSampler
+
+
+def make_batch(image, mask, device):
+    image = np.array(Image.open(image).convert("RGB"))
+    image = image.astype(np.float32)/255.0
+    image = image[None].transpose(0,3,1,2)
+    image = torch.from_numpy(image)
+
+    mask = np.array(Image.open(mask).convert("L"))
+    mask = mask.astype(np.float32)/255.0
+    mask = mask[None,None]
+    mask[mask < 0.5] = 0
+    mask[mask >= 0.5] = 1
+    mask = torch.from_numpy(mask)
+
+    masked_image = (1-mask)*image
+
+    batch = {"image": image, "mask": mask, "masked_image": masked_image}
+    for k in batch:
+        batch[k] = batch[k].to(device=device)
+        batch[k] = batch[k]*2.0-1.0
+    return batch
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--indir",
+        type=str,
+        nargs="?",
+        help="dir containing image-mask pairs (`example.png` and `example_mask.png`)",
+    )
+    parser.add_argument(
+        "--outdir",
+        type=str,
+        nargs="?",
+        help="dir to write results to",
+    )
+    parser.add_argument(
+        "--steps",
+        type=int,
+        default=50,
+        help="number of ddim sampling steps",
+    )
+    opt = parser.parse_args()
+
+    masks = sorted(glob.glob(os.path.join(opt.indir, "*_mask.png")))
+    images = [x.replace("_mask.png", ".png") for x in masks]
+    print(f"Found {len(masks)} inputs.")
+
+    config = OmegaConf.load("models/ldm/inpainting_big/config.yaml")
+    model = instantiate_from_config(config.model)
+    model.load_state_dict(torch.load("models/ldm/inpainting_big/last.ckpt")["state_dict"],
+                          strict=False)
+
+    device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")
+    model = model.to(device)
+    sampler = DDIMSampler(model)
+
+    os.makedirs(opt.outdir, exist_ok=True)
+    with torch.no_grad():
+        with model.ema_scope():
+            for image, mask in tqdm(zip(images, masks)):
+                outpath = os.path.join(opt.outdir, os.path.split(image)[1])
+                batch = make_batch(image, mask, device=device)
+
+                # encode masked image and concat downsampled mask
+                c = model.cond_stage_model.encode(batch["masked_image"])
+                cc = torch.nn.functional.interpolate(batch["mask"],
+                                                     size=c.shape[-2:])
+                c = torch.cat((c, cc), dim=1)
+
+                shape = (c.shape[1]-1,)+c.shape[2:]
+                samples_ddim, _ = sampler.sample(S=opt.steps,
+                                                 conditioning=c,
+                                                 batch_size=c.shape[0],
+                                                 shape=shape,
+                                                 verbose=False)
+                x_samples_ddim = model.decode_first_stage(samples_ddim)
+
+                image = torch.clamp((batch["image"]+1.0)/2.0,
+                                    min=0.0, max=1.0)
+                mask = torch.clamp((batch["mask"]+1.0)/2.0,
+                                   min=0.0, max=1.0)
+                predicted_image = torch.clamp((x_samples_ddim+1.0)/2.0,
+                                              min=0.0, max=1.0)
+
+                inpainted = (1-mask)*image+mask*predicted_image
+                inpainted = inpainted.cpu().numpy().transpose(0,2,3,1)[0]*255
+                Image.fromarray(inpainted.astype(np.uint8)).save(outpath)

examples/tutorial/stable_diffusion/scripts/knn2img.py

+import argparse, os, sys, glob
+import clip
+import torch
+import torch.nn as nn
+import numpy as np
+from omegaconf import OmegaConf
+from PIL import Image
+from tqdm import tqdm, trange
+from itertools import islice
+from einops import rearrange, repeat
+from torchvision.utils import make_grid
+import scann
+import time
+from multiprocessing import cpu_count
+
+from ldm.util import instantiate_from_config, parallel_data_prefetch
+from ldm.models.diffusion.ddim import DDIMSampler
+from ldm.models.diffusion.plms import PLMSSampler
+from ldm.modules.encoders.modules import FrozenClipImageEmbedder, FrozenCLIPTextEmbedder
+
+DATABASES = [
+    "openimages",
+    "artbench-art_nouveau",
+    "artbench-baroque",
+    "artbench-expressionism",
+    "artbench-impressionism",
+    "artbench-post_impressionism",
+    "artbench-realism",
+    "artbench-romanticism",
+    "artbench-renaissance",
+    "artbench-surrealism",
+    "artbench-ukiyo_e",
+]
+
+
+def chunk(it, size):
+    it = iter(it)
+    return iter(lambda: tuple(islice(it, size)), ())
+
+
+def load_model_from_config(config, ckpt, verbose=False):
+    print(f"Loading model from {ckpt}")
+    pl_sd = torch.load(ckpt, map_location="cpu")
+    if "global_step" in pl_sd:
+        print(f"Global Step: {pl_sd['global_step']}")
+    sd = pl_sd["state_dict"]
+    model = instantiate_from_config(config.model)
+    m, u = model.load_state_dict(sd, strict=False)
+    if len(m) > 0 and verbose:
+        print("missing keys:")
+        print(m)
+    if len(u) > 0 and verbose:
+        print("unexpected keys:")
+        print(u)
+
+    model.cuda()
+    model.eval()
+    return model
+
+
+class Searcher(object):
+    def __init__(self, database, retriever_version='ViT-L/14'):
+        assert database in DATABASES
+        # self.database = self.load_database(database)
+        self.database_name = database
+        self.searcher_savedir = f'data/rdm/searchers/{self.database_name}'
+        self.database_path = f'data/rdm/retrieval_databases/{self.database_name}'
+        self.retriever = self.load_retriever(version=retriever_version)
+        self.database = {'embedding': [],
+                         'img_id': [],
+                         'patch_coords': []}
+        self.load_database()
+        self.load_searcher()
+
+    def train_searcher(self, k,
+                       metric='dot_product',
+                       searcher_savedir=None):
+
+        print('Start training searcher')
+        searcher = scann.scann_ops_pybind.builder(self.database['embedding'] /
+                                                  np.linalg.norm(self.database['embedding'], axis=1)[:, np.newaxis],
+                                                  k, metric)
+        self.searcher = searcher.score_brute_force().build()
+        print('Finish training searcher')
+
+        if searcher_savedir is not None:
+            print(f'Save trained searcher under "{searcher_savedir}"')
+            os.makedirs(searcher_savedir, exist_ok=True)
+            self.searcher.serialize(searcher_savedir)
+
+    def load_single_file(self, saved_embeddings):
+        compressed = np.load(saved_embeddings)
+        self.database = {key: compressed[key] for key in compressed.files}
+        print('Finished loading of clip embeddings.')
+
+    def load_multi_files(self, data_archive):
+        out_data = {key: [] for key in self.database}
+        for d in tqdm(data_archive, desc=f'Loading datapool from {len(data_archive)} individual files.'):
+            for key in d.files:
+                out_data[key].append(d[key])
+
+        return out_data
+
+    def load_database(self):
+
+        print(f'Load saved patch embedding from "{self.database_path}"')
+        file_content = glob.glob(os.path.join(self.database_path, '*.npz'))
+
+        if len(file_content) == 1:
+            self.load_single_file(file_content[0])
+        elif len(file_content) > 1:
+            data = [np.load(f) for f in file_content]
+            prefetched_data = parallel_data_prefetch(self.load_multi_files, data,
+                                                     n_proc=min(len(data), cpu_count()), target_data_type='dict')
+
+            self.database = {key: np.concatenate([od[key] for od in prefetched_data], axis=1)[0] for key in
+                             self.database}
+        else:
+            raise ValueError(f'No npz-files in specified path "{self.database_path}" is this directory existing?')
+
+        print(f'Finished loading of retrieval database of length {self.database["embedding"].shape[0]}.')
+
+    def load_retriever(self, version='ViT-L/14', ):
+        model = FrozenClipImageEmbedder(model=version)
+        if torch.cuda.is_available():
+            model.cuda()
+        model.eval()
+        return model
+
+    def load_searcher(self):
+        print(f'load searcher for database {self.database_name} from {self.searcher_savedir}')
+        self.searcher = scann.scann_ops_pybind.load_searcher(self.searcher_savedir)
+        print('Finished loading searcher.')
+
+    def search(self, x, k):
+        if self.searcher is None and self.database['embedding'].shape[0] < 2e4:
+            self.train_searcher(k)   # quickly fit searcher on the fly for small databases
+        assert self.searcher is not None, 'Cannot search with uninitialized searcher'
+        if isinstance(x, torch.Tensor):
+            x = x.detach().cpu().numpy()
+        if len(x.shape) == 3:
+            x = x[:, 0]
+        query_embeddings = x / np.linalg.norm(x, axis=1)[:, np.newaxis]
+
+        start = time.time()
+        nns, distances = self.searcher.search_batched(query_embeddings, final_num_neighbors=k)
+        end = time.time()
+
+        out_embeddings = self.database['embedding'][nns]
+        out_img_ids = self.database['img_id'][nns]
+        out_pc = self.database['patch_coords'][nns]
+
+        out = {'nn_embeddings': out_embeddings / np.linalg.norm(out_embeddings, axis=-1)[..., np.newaxis],
+               'img_ids': out_img_ids,
+               'patch_coords': out_pc,
+               'queries': x,
+               'exec_time': end - start,
+               'nns': nns,
+               'q_embeddings': query_embeddings}
+
+        return out
+
+    def __call__(self, x, n):
+        return self.search(x, n)
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    # TODO: add n_neighbors and modes (text-only, text-image-retrieval, image-image retrieval etc)
+    # TODO: add 'image variation' mode when knn=0 but a single image is given instead of a text prompt?
+    parser.add_argument(
+        "--prompt",
+        type=str,
+        nargs="?",
+        default="a painting of a virus monster playing guitar",
+        help="the prompt to render"
+    )
+
+    parser.add_argument(
+        "--outdir",
+        type=str,
+        nargs="?",
+        help="dir to write results to",
+        default="outputs/txt2img-samples"
+    )
+
+    parser.add_argument(
+        "--skip_grid",
+        action='store_true',
+        help="do not save a grid, only individual samples. Helpful when evaluating lots of samples",
+    )
+
+    parser.add_argument(
+        "--ddim_steps",
+        type=int,
+        default=50,
+        help="number of ddim sampling steps",
+    )
+
+    parser.add_argument(
+        "--n_repeat",
+        type=int,
+        default=1,
+        help="number of repeats in CLIP latent space",
+    )
+
+    parser.add_argument(
+        "--plms",
+        action='store_true',
+        help="use plms sampling",
+    )
+
+    parser.add_argument(
+        "--ddim_eta",
+        type=float,
+        default=0.0,
+        help="ddim eta (eta=0.0 corresponds to deterministic sampling",
+    )
+    parser.add_argument(
+        "--n_iter",
+        type=int,
+        default=1,
+        help="sample this often",
+    )
+
+    parser.add_argument(
+        "--H",
+        type=int,
+        default=768,
+        help="image height, in pixel space",
+    )
+
+    parser.add_argument(
+        "--W",
+        type=int,
+        default=768,
+        help="image width, in pixel space",
+    )
+
+    parser.add_argument(
+        "--n_samples",
+        type=int,
+        default=3,
+        help="how many samples to produce for each given prompt. A.k.a batch size",
+    )
+
+    parser.add_argument(
+        "--n_rows",
+        type=int,
+        default=0,
+        help="rows in the grid (default: n_samples)",
+    )
+
+    parser.add_argument(
+        "--scale",
+        type=float,
+        default=5.0,
+        help="unconditional guidance scale: eps = eps(x, empty) + scale * (eps(x, cond) - eps(x, empty))",
+    )
+
+    parser.add_argument(
+        "--from-file",
+        type=str,
+        help="if specified, load prompts from this file",
+    )
+
+    parser.add_argument(
+        "--config",
+        type=str,
+        default="configs/retrieval-augmented-diffusion/768x768.yaml",
+        help="path to config which constructs model",
+    )
+
+    parser.add_argument(
+        "--ckpt",
+        type=str,
+        default="models/rdm/rdm768x768/model.ckpt",
+        help="path to checkpoint of model",
+    )
+
+    parser.add_argument(
+        "--clip_type",
+        type=str,
+        default="ViT-L/14",
+        help="which CLIP model to use for retrieval and NN encoding",
+    )
+    parser.add_argument(
+        "--database",
+        type=str,
+        default='artbench-surrealism',
+        choices=DATABASES,
+        help="The database used for the search, only applied when --use_neighbors=True",
+    )
+    parser.add_argument(
+        "--use_neighbors",
+        default=False,
+        action='store_true',
+        help="Include neighbors in addition to text prompt for conditioning",
+    )
+    parser.add_argument(
+        "--knn",
+        default=10,
+        type=int,
+        help="The number of included neighbors, only applied when --use_neighbors=True",
+    )
+
+    opt = parser.parse_args()
+
+    config = OmegaConf.load(f"{opt.config}")
+    model = load_model_from_config(config, f"{opt.ckpt}")
+
+    device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")
+    model = model.to(device)
+
+    clip_text_encoder = FrozenCLIPTextEmbedder(opt.clip_type).to(device)
+
+    if opt.plms:
+        sampler = PLMSSampler(model)
+    else:
+        sampler = DDIMSampler(model)
+
+    os.makedirs(opt.outdir, exist_ok=True)
+    outpath = opt.outdir
+
+    batch_size = opt.n_samples
+    n_rows = opt.n_rows if opt.n_rows > 0 else batch_size
+    if not opt.from_file:
+        prompt = opt.prompt
+        assert prompt is not None
+        data = [batch_size * [prompt]]
+
+    else:
+        print(f"reading prompts from {opt.from_file}")
+        with open(opt.from_file, "r") as f:
+            data = f.read().splitlines()
+            data = list(chunk(data, batch_size))
+
+    sample_path = os.path.join(outpath, "samples")
+    os.makedirs(sample_path, exist_ok=True)
+    base_count = len(os.listdir(sample_path))
+    grid_count = len(os.listdir(outpath)) - 1
+
+    print(f"sampling scale for cfg is {opt.scale:.2f}")
+
+    searcher = None
+    if opt.use_neighbors:
+        searcher = Searcher(opt.database)
+
+    with torch.no_grad():
+        with model.ema_scope():
+            for n in trange(opt.n_iter, desc="Sampling"):
+                all_samples = list()
+                for prompts in tqdm(data, desc="data"):
+                    print("sampling prompts:", prompts)
+                    if isinstance(prompts, tuple):
+                        prompts = list(prompts)
+                    c = clip_text_encoder.encode(prompts)
+                    uc = None
+                    if searcher is not None:
+                        nn_dict = searcher(c, opt.knn)
+                        c = torch.cat([c, torch.from_numpy(nn_dict['nn_embeddings']).cuda()], dim=1)
+                    if opt.scale != 1.0:
+                        uc = torch.zeros_like(c)
+                    if isinstance(prompts, tuple):
+                        prompts = list(prompts)
+                    shape = [16, opt.H // 16, opt.W // 16]  # note: currently hardcoded for f16 model
+                    samples_ddim, _ = sampler.sample(S=opt.ddim_steps,
+                                                     conditioning=c,
+                                                     batch_size=c.shape[0],
+                                                     shape=shape,
+                                                     verbose=False,
+                                                     unconditional_guidance_scale=opt.scale,
+                                                     unconditional_conditioning=uc,
+                                                     eta=opt.ddim_eta,
+                                                     )
+
+                    x_samples_ddim = model.decode_first_stage(samples_ddim)
+                    x_samples_ddim = torch.clamp((x_samples_ddim + 1.0) / 2.0, min=0.0, max=1.0)
+
+                    for x_sample in x_samples_ddim:
+                        x_sample = 255. * rearrange(x_sample.cpu().numpy(), 'c h w -> h w c')
+                        Image.fromarray(x_sample.astype(np.uint8)).save(
+                            os.path.join(sample_path, f"{base_count:05}.png"))
+                        base_count += 1
+                    all_samples.append(x_samples_ddim)
+
+                if not opt.skip_grid:
+                    # additionally, save as grid
+                    grid = torch.stack(all_samples, 0)
+                    grid = rearrange(grid, 'n b c h w -> (n b) c h w')
+                    grid = make_grid(grid, nrow=n_rows)
+
+                    # to image
+                    grid = 255. * rearrange(grid, 'c h w -> h w c').cpu().numpy()
+                    Image.fromarray(grid.astype(np.uint8)).save(os.path.join(outpath, f'grid-{grid_count:04}.png'))
+                    grid_count += 1
+
+    print(f"Your samples are ready and waiting for you here: \n{outpath} \nEnjoy.")

examples/tutorial/stable_diffusion/scripts/sample_diffusion.py

+import argparse, os, sys, glob, datetime, yaml
+import torch
+import time
+import numpy as np
+from tqdm import trange
+
+from omegaconf import OmegaConf
+from PIL import Image
+
+from ldm.models.diffusion.ddim import DDIMSampler
+from ldm.util import instantiate_from_config
+
+rescale = lambda x: (x + 1.) / 2.
+
+def custom_to_pil(x):
+    x = x.detach().cpu()
+    x = torch.clamp(x, -1., 1.)
+    x = (x + 1.) / 2.
+    x = x.permute(1, 2, 0).numpy()
+    x = (255 * x).astype(np.uint8)
+    x = Image.fromarray(x)
+    if not x.mode == "RGB":
+        x = x.convert("RGB")
+    return x
+
+
+def custom_to_np(x):
+    # saves the batch in adm style as in https://github.com/openai/guided-diffusion/blob/main/scripts/image_sample.py
+    sample = x.detach().cpu()
+    sample = ((sample + 1) * 127.5).clamp(0, 255).to(torch.uint8)
+    sample = sample.permute(0, 2, 3, 1)
+    sample = sample.contiguous()
+    return sample
+
+
+def logs2pil(logs, keys=["sample"]):
+    imgs = dict()
+    for k in logs:
+        try:
+            if len(logs[k].shape) == 4:
+                img = custom_to_pil(logs[k][0, ...])
+            elif len(logs[k].shape) == 3:
+                img = custom_to_pil(logs[k])
+            else:
+                print(f"Unknown format for key {k}. ")
+                img = None
+        except:
+            img = None
+        imgs[k] = img
+    return imgs
+
+
+@torch.no_grad()
+def convsample(model, shape, return_intermediates=True,
+               verbose=True,
+               make_prog_row=False):
+
+
+    if not make_prog_row:
+        return model.p_sample_loop(None, shape,
+                                   return_intermediates=return_intermediates, verbose=verbose)
+    else:
+        return model.progressive_denoising(
+            None, shape, verbose=True
+        )
+
+
+@torch.no_grad()
+def convsample_ddim(model, steps, shape, eta=1.0
+                    ):
+    ddim = DDIMSampler(model)
+    bs = shape[0]
+    shape = shape[1:]
+    samples, intermediates = ddim.sample(steps, batch_size=bs, shape=shape, eta=eta, verbose=False,)
+    return samples, intermediates
+
+
+@torch.no_grad()
+def make_convolutional_sample(model, batch_size, vanilla=False, custom_steps=None, eta=1.0,):
+
+
+    log = dict()
+
+    shape = [batch_size,
+             model.model.diffusion_model.in_channels,
+             model.model.diffusion_model.image_size,
+             model.model.diffusion_model.image_size]
+
+    with model.ema_scope("Plotting"):
+        t0 = time.time()
+        if vanilla:
+            sample, progrow = convsample(model, shape,
+                                         make_prog_row=True)
+        else:
+            sample, intermediates = convsample_ddim(model,  steps=custom_steps, shape=shape,
+                                                    eta=eta)
+
+        t1 = time.time()
+
+    x_sample = model.decode_first_stage(sample)
+
+    log["sample"] = x_sample
+    log["time"] = t1 - t0
+    log['throughput'] = sample.shape[0] / (t1 - t0)
+    print(f'Throughput for this batch: {log["throughput"]}')
+    return log
+
+def run(model, logdir, batch_size=50, vanilla=False, custom_steps=None, eta=None, n_samples=50000, nplog=None):
+    if vanilla:
+        print(f'Using Vanilla DDPM sampling with {model.num_timesteps} sampling steps.')
+    else:
+        print(f'Using DDIM sampling with {custom_steps} sampling steps and eta={eta}')
+
+
+    tstart = time.time()
+    n_saved = len(glob.glob(os.path.join(logdir,'*.png')))-1
+    # path = logdir
+    if model.cond_stage_model is None:
+        all_images = []
+
+        print(f"Running unconditional sampling for {n_samples} samples")
+        for _ in trange(n_samples // batch_size, desc="Sampling Batches (unconditional)"):
+            logs = make_convolutional_sample(model, batch_size=batch_size,
+                                             vanilla=vanilla, custom_steps=custom_steps,
+                                             eta=eta)
+            n_saved = save_logs(logs, logdir, n_saved=n_saved, key="sample")
+            all_images.extend([custom_to_np(logs["sample"])])
+            if n_saved >= n_samples:
+                print(f'Finish after generating {n_saved} samples')
+                break
+        all_img = np.concatenate(all_images, axis=0)
+        all_img = all_img[:n_samples]
+        shape_str = "x".join([str(x) for x in all_img.shape])
+        nppath = os.path.join(nplog, f"{shape_str}-samples.npz")
+        np.savez(nppath, all_img)
+
+    else:
+       raise NotImplementedError('Currently only sampling for unconditional models supported.')
+
+    print(f"sampling of {n_saved} images finished in {(time.time() - tstart) / 60.:.2f} minutes.")
+
+
+def save_logs(logs, path, n_saved=0, key="sample", np_path=None):
+    for k in logs:
+        if k == key:
+            batch = logs[key]
+            if np_path is None:
+                for x in batch:
+                    img = custom_to_pil(x)
+                    imgpath = os.path.join(path, f"{key}_{n_saved:06}.png")
+                    img.save(imgpath)
+                    n_saved += 1
+            else:
+                npbatch = custom_to_np(batch)
+                shape_str = "x".join([str(x) for x in npbatch.shape])
+                nppath = os.path.join(np_path, f"{n_saved}-{shape_str}-samples.npz")
+                np.savez(nppath, npbatch)
+                n_saved += npbatch.shape[0]
+    return n_saved
+
+
+def get_parser():
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "-r",
+        "--resume",
+        type=str,
+        nargs="?",
+        help="load from logdir or checkpoint in logdir",
+    )
+    parser.add_argument(
+        "-n",
+        "--n_samples",
+        type=int,
+        nargs="?",
+        help="number of samples to draw",
+        default=50000
+    )
+    parser.add_argument(
+        "-e",
+        "--eta",
+        type=float,
+        nargs="?",
+        help="eta for ddim sampling (0.0 yields deterministic sampling)",
+        default=1.0
+    )
+    parser.add_argument(
+        "-v",
+        "--vanilla_sample",
+        default=False,
+        action='store_true',
+        help="vanilla sampling (default option is DDIM sampling)?",
+    )
+    parser.add_argument(
+        "-l",
+        "--logdir",
+        type=str,
+        nargs="?",
+        help="extra logdir",
+        default="none"
+    )
+    parser.add_argument(
+        "-c",
+        "--custom_steps",
+        type=int,
+        nargs="?",
+        help="number of steps for ddim and fastdpm sampling",
+        default=50
+    )
+    parser.add_argument(
+        "--batch_size",
+        type=int,
+        nargs="?",
+        help="the bs",
+        default=10
+    )
+    return parser
+
+
+def load_model_from_config(config, sd):
+    model = instantiate_from_config(config)
+    model.load_state_dict(sd,strict=False)
+    model.cuda()
+    model.eval()
+    return model
+
+
+def load_model(config, ckpt, gpu, eval_mode):
+    if ckpt:
+        print(f"Loading model from {ckpt}")
+        pl_sd = torch.load(ckpt, map_location="cpu")
+        global_step = pl_sd["global_step"]
+    else:
+        pl_sd = {"state_dict": None}
+        global_step = None
+    model = load_model_from_config(config.model,
+                                   pl_sd["state_dict"])
+
+    return model, global_step
+
+
+if __name__ == "__main__":
+    now = datetime.datetime.now().strftime("%Y-%m-%d-%H-%M-%S")
+    sys.path.append(os.getcwd())
+    command = " ".join(sys.argv)
+
+    parser = get_parser()
+    opt, unknown = parser.parse_known_args()
+    ckpt = None
+
+    if not os.path.exists(opt.resume):
+        raise ValueError("Cannot find {}".format(opt.resume))
+    if os.path.isfile(opt.resume):
+        # paths = opt.resume.split("/")
+        try:
+            logdir = '/'.join(opt.resume.split('/')[:-1])
+            # idx = len(paths)-paths[::-1].index("logs")+1
+            print(f'Logdir is {logdir}')
+        except ValueError:
+            paths = opt.resume.split("/")
+            idx = -2  # take a guess: path/to/logdir/checkpoints/model.ckpt
+            logdir = "/".join(paths[:idx])
+        ckpt = opt.resume
+    else:
+        assert os.path.isdir(opt.resume), f"{opt.resume} is not a directory"
+        logdir = opt.resume.rstrip("/")
+        ckpt = os.path.join(logdir, "model.ckpt")
+
+    base_configs = sorted(glob.glob(os.path.join(logdir, "config.yaml")))
+    opt.base = base_configs
+
+    configs = [OmegaConf.load(cfg) for cfg in opt.base]
+    cli = OmegaConf.from_dotlist(unknown)
+    config = OmegaConf.merge(*configs, cli)
+
+    gpu = True
+    eval_mode = True
+
+    if opt.logdir != "none":
+        locallog = logdir.split(os.sep)[-1]
+        if locallog == "": locallog = logdir.split(os.sep)[-2]
+        print(f"Switching logdir from '{logdir}' to '{os.path.join(opt.logdir, locallog)}'")
+        logdir = os.path.join(opt.logdir, locallog)
+
+    print(config)
+
+    model, global_step = load_model(config, ckpt, gpu, eval_mode)
+    print(f"global step: {global_step}")
+    print(75 * "=")
+    print("logging to:")
+    logdir = os.path.join(logdir, "samples", f"{global_step:08}", now)
+    imglogdir = os.path.join(logdir, "img")
+    numpylogdir = os.path.join(logdir, "numpy")
+
+    os.makedirs(imglogdir)
+    os.makedirs(numpylogdir)
+    print(logdir)
+    print(75 * "=")
+
+    # write config out
+    sampling_file = os.path.join(logdir, "sampling_config.yaml")
+    sampling_conf = vars(opt)
+
+    with open(sampling_file, 'w') as f:
+        yaml.dump(sampling_conf, f, default_flow_style=False)
+    print(sampling_conf)
+
+
+    run(model, imglogdir, eta=opt.eta,
+        vanilla=opt.vanilla_sample,  n_samples=opt.n_samples, custom_steps=opt.custom_steps,
+        batch_size=opt.batch_size, nplog=numpylogdir)
+
+    print("done.")

examples/tutorial/stable_diffusion/scripts/tests/test_checkpoint.py

+import os
+import sys
+from copy import deepcopy
+
+import yaml
+from datetime import datetime
+
+from diffusers import StableDiffusionPipeline
+import torch
+from ldm.util import instantiate_from_config
+from main import get_parser
+
+if __name__ == "__main__":
+    with torch.no_grad():
+        yaml_path = "../../train_colossalai.yaml"
+        with open(yaml_path, 'r', encoding='utf-8') as f:
+            config = f.read()
+        base_config = yaml.load(config, Loader=yaml.FullLoader)
+        unet_config = base_config['model']['params']['unet_config']
+        diffusion_model = instantiate_from_config(unet_config).to("cuda:0")
+
+        pipe = StableDiffusionPipeline.from_pretrained(
+            "/data/scratch/diffuser/stable-diffusion-v1-4"
+        ).to("cuda:0")
+        dif_model_2 = pipe.unet
+
+        random_input_ = torch.rand((4, 4, 32, 32)).to("cuda:0")
+        random_input_2 = torch.clone(random_input_).to("cuda:0")
+        time_stamp = torch.randint(20, (4,)).to("cuda:0")
+        time_stamp2 = torch.clone(time_stamp).to("cuda:0")
+        context_ = torch.rand((4, 77, 768)).to("cuda:0")
+        context_2 = torch.clone(context_).to("cuda:0")
+
+        out_1 = diffusion_model(random_input_, time_stamp, context_)
+        out_2 = dif_model_2(random_input_2, time_stamp2, context_2)
+        print(out_1.shape)
+        print(out_2['sample'].shape)
\ No newline at end of file

examples/tutorial/stable_diffusion/scripts/tests/test_watermark.py

+import cv2
+import fire
+from imwatermark import WatermarkDecoder
+
+
+def testit(img_path):
+    bgr = cv2.imread(img_path)
+    decoder = WatermarkDecoder('bytes', 136)
+    watermark = decoder.decode(bgr, 'dwtDct')
+    try:
+        dec = watermark.decode('utf-8')
+    except:
+        dec = "null"
+    print(dec)
+
+
+if __name__ == "__main__":
+    fire.Fire(testit)
\ No newline at end of file

examples/tutorial/stable_diffusion/scripts/train_searcher.py

+import os, sys
+import numpy as np
+import scann
+import argparse
+import glob
+from multiprocessing import cpu_count
+from tqdm import tqdm
+
+from ldm.util import parallel_data_prefetch
+
+
+def search_bruteforce(searcher):
+    return searcher.score_brute_force().build()
+
+
+def search_partioned_ah(searcher, dims_per_block, aiq_threshold, reorder_k,
+                        partioning_trainsize, num_leaves, num_leaves_to_search):
+    return searcher.tree(num_leaves=num_leaves,
+                         num_leaves_to_search=num_leaves_to_search,
+                         training_sample_size=partioning_trainsize). \
+        score_ah(dims_per_block, anisotropic_quantization_threshold=aiq_threshold).reorder(reorder_k).build()
+
+
+def search_ah(searcher, dims_per_block, aiq_threshold, reorder_k):
+    return searcher.score_ah(dims_per_block, anisotropic_quantization_threshold=aiq_threshold).reorder(
+        reorder_k).build()
+
+def load_datapool(dpath):
+
+
+    def load_single_file(saved_embeddings):
+        compressed = np.load(saved_embeddings)
+        database = {key: compressed[key] for key in compressed.files}
+        return database
+
+    def load_multi_files(data_archive):
+        database = {key: [] for key in data_archive[0].files}
+        for d in tqdm(data_archive, desc=f'Loading datapool from {len(data_archive)} individual files.'):
+            for key in d.files:
+                database[key].append(d[key])
+
+        return database
+
+    print(f'Load saved patch embedding from "{dpath}"')
+    file_content = glob.glob(os.path.join(dpath, '*.npz'))
+
+    if len(file_content) == 1:
+        data_pool = load_single_file(file_content[0])
+    elif len(file_content) > 1:
+        data = [np.load(f) for f in file_content]
+        prefetched_data = parallel_data_prefetch(load_multi_files, data,
+                                                 n_proc=min(len(data), cpu_count()), target_data_type='dict')
+
+        data_pool = {key: np.concatenate([od[key] for od in prefetched_data], axis=1)[0] for key in prefetched_data[0].keys()}
+    else:
+        raise ValueError(f'No npz-files in specified path "{dpath}" is this directory existing?')
+
+    print(f'Finished loading of retrieval database of length {data_pool["embedding"].shape[0]}.')
+    return data_pool
+
+
+def train_searcher(opt,
+                   metric='dot_product',
+                   partioning_trainsize=None,
+                   reorder_k=None,
+                   # todo tune
+                   aiq_thld=0.2,
+                   dims_per_block=2,
+                   num_leaves=None,
+                   num_leaves_to_search=None,):
+
+    data_pool = load_datapool(opt.database)
+    k = opt.knn
+
+    if not reorder_k:
+        reorder_k = 2 * k
+
+    # normalize
+    # embeddings =
+    searcher = scann.scann_ops_pybind.builder(data_pool['embedding'] / np.linalg.norm(data_pool['embedding'], axis=1)[:, np.newaxis], k, metric)
+    pool_size = data_pool['embedding'].shape[0]
+
+    print(*(['#'] * 100))
+    print('Initializing scaNN searcher with the following values:')
+    print(f'k: {k}')
+    print(f'metric: {metric}')
+    print(f'reorder_k: {reorder_k}')
+    print(f'anisotropic_quantization_threshold: {aiq_thld}')
+    print(f'dims_per_block: {dims_per_block}')
+    print(*(['#'] * 100))
+    print('Start training searcher....')
+    print(f'N samples in pool is {pool_size}')
+
+    # this reflects the recommended design choices proposed at
+    # https://github.com/google-research/google-research/blob/aca5f2e44e301af172590bb8e65711f0c9ee0cfd/scann/docs/algorithms.md
+    if pool_size < 2e4:
+        print('Using brute force search.')
+        searcher = search_bruteforce(searcher)
+    elif 2e4 <= pool_size and pool_size < 1e5:
+        print('Using asymmetric hashing search and reordering.')
+        searcher = search_ah(searcher, dims_per_block, aiq_thld, reorder_k)
+    else:
+        print('Using using partioning, asymmetric hashing search and reordering.')
+
+        if not partioning_trainsize:
+            partioning_trainsize = data_pool['embedding'].shape[0] // 10
+        if not num_leaves:
+            num_leaves = int(np.sqrt(pool_size))
+
+        if not num_leaves_to_search:
+            num_leaves_to_search = max(num_leaves // 20, 1)
+
+        print('Partitioning params:')
+        print(f'num_leaves: {num_leaves}')
+        print(f'num_leaves_to_search: {num_leaves_to_search}')
+        # self.searcher = self.search_ah(searcher, dims_per_block, aiq_thld, reorder_k)
+        searcher = search_partioned_ah(searcher, dims_per_block, aiq_thld, reorder_k,
+                                                 partioning_trainsize, num_leaves, num_leaves_to_search)
+
+    print('Finish training searcher')
+    searcher_savedir = opt.target_path
+    os.makedirs(searcher_savedir, exist_ok=True)
+    searcher.serialize(searcher_savedir)
+    print(f'Saved trained searcher under "{searcher_savedir}"')
+
+if __name__ == '__main__':
+    sys.path.append(os.getcwd())
+    parser = argparse.ArgumentParser()
+    parser.add_argument('--database',
+                        '-d',
+                        default='data/rdm/retrieval_databases/openimages',
+                        type=str,
+                        help='path to folder containing the clip feature of the database')
+    parser.add_argument('--target_path',
+                        '-t',
+                        default='data/rdm/searchers/openimages',
+                        type=str,
+                        help='path to the target folder where the searcher shall be stored.')
+    parser.add_argument('--knn',
+                        '-k',
+                        default=20,
+                        type=int,
+                        help='number of nearest neighbors, for which the searcher shall be optimized')
+
+    opt, _  = parser.parse_known_args()
+
+    train_searcher(opt,)
\ No newline at end of file

examples/tutorial/stable_diffusion/scripts/txt2img.py

+import argparse, os, sys, glob
+import cv2
+import torch
+import numpy as np
+from omegaconf import OmegaConf
+from PIL import Image
+from tqdm import tqdm, trange
+from imwatermark import WatermarkEncoder
+from itertools import islice
+from einops import rearrange
+from torchvision.utils import make_grid
+import time
+from pytorch_lightning import seed_everything
+from torch import autocast
+from contextlib import contextmanager, nullcontext
+
+from ldm.util import instantiate_from_config
+from ldm.models.diffusion.ddim import DDIMSampler
+from ldm.models.diffusion.plms import PLMSSampler
+
+from diffusers.pipelines.stable_diffusion.safety_checker import StableDiffusionSafetyChecker
+from transformers import AutoFeatureExtractor
+
+
+# load safety model
+safety_model_id = "CompVis/stable-diffusion-safety-checker"
+safety_feature_extractor = AutoFeatureExtractor.from_pretrained(safety_model_id)
+safety_checker = StableDiffusionSafetyChecker.from_pretrained(safety_model_id)
+
+
+def chunk(it, size):
+    it = iter(it)
+    return iter(lambda: tuple(islice(it, size)), ())
+
+
+def numpy_to_pil(images):
+    """
+    Convert a numpy image or a batch of images to a PIL image.
+    """
+    if images.ndim == 3:
+        images = images[None, ...]
+    images = (images * 255).round().astype("uint8")
+    pil_images = [Image.fromarray(image) for image in images]
+
+    return pil_images
+
+
+def load_model_from_config(config, ckpt, verbose=False):
+    print(f"Loading model from {ckpt}")
+    pl_sd = torch.load(ckpt, map_location="cpu")
+    if "global_step" in pl_sd:
+        print(f"Global Step: {pl_sd['global_step']}")
+    sd = pl_sd["state_dict"]
+    model = instantiate_from_config(config.model)
+    m, u = model.load_state_dict(sd, strict=False)
+    if len(m) > 0 and verbose:
+        print("missing keys:")
+        print(m)
+    if len(u) > 0 and verbose:
+        print("unexpected keys:")
+        print(u)
+
+    model.cuda()
+    model.eval()
+    return model
+
+
+def put_watermark(img, wm_encoder=None):
+    if wm_encoder is not None:
+        img = cv2.cvtColor(np.array(img), cv2.COLOR_RGB2BGR)
+        img = wm_encoder.encode(img, 'dwtDct')
+        img = Image.fromarray(img[:, :, ::-1])
+    return img
+
+
+def load_replacement(x):
+    try:
+        hwc = x.shape
+        y = Image.open("assets/rick.jpeg").convert("RGB").resize((hwc[1], hwc[0]))
+        y = (np.array(y)/255.0).astype(x.dtype)
+        assert y.shape == x.shape
+        return y
+    except Exception:
+        return x
+
+
+def check_safety(x_image):
+    safety_checker_input = safety_feature_extractor(numpy_to_pil(x_image), return_tensors="pt")
+    x_checked_image, has_nsfw_concept = safety_checker(images=x_image, clip_input=safety_checker_input.pixel_values)
+    assert x_checked_image.shape[0] == len(has_nsfw_concept)
+    for i in range(len(has_nsfw_concept)):
+        if has_nsfw_concept[i]:
+            x_checked_image[i] = load_replacement(x_checked_image[i])
+    return x_checked_image, has_nsfw_concept
+
+
+def main():
+    parser = argparse.ArgumentParser()
+
+    parser.add_argument(
+        "--prompt",
+        type=str,
+        nargs="?",
+        default="a painting of a virus monster playing guitar",
+        help="the prompt to render"
+    )
+    parser.add_argument(
+        "--outdir",
+        type=str,
+        nargs="?",
+        help="dir to write results to",
+        default="outputs/txt2img-samples"
+    )
+    parser.add_argument(
+        "--skip_grid",
+        action='store_true',
+        help="do not save a grid, only individual samples. Helpful when evaluating lots of samples",
+    )
+    parser.add_argument(
+        "--skip_save",
+        action='store_true',
+        help="do not save individual samples. For speed measurements.",
+    )
+    parser.add_argument(
+        "--ddim_steps",
+        type=int,
+        default=50,
+        help="number of ddim sampling steps",
+    )
+    parser.add_argument(
+        "--plms",
+        action='store_true',
+        help="use plms sampling",
+    )
+    parser.add_argument(
+        "--laion400m",
+        action='store_true',
+        help="uses the LAION400M model",
+    )
+    parser.add_argument(
+        "--fixed_code",
+        action='store_true',
+        help="if enabled, uses the same starting code across samples ",
+    )
+    parser.add_argument(
+        "--ddim_eta",
+        type=float,
+        default=0.0,
+        help="ddim eta (eta=0.0 corresponds to deterministic sampling",
+    )
+    parser.add_argument(
+        "--n_iter",
+        type=int,
+        default=2,
+        help="sample this often",
+    )
+    parser.add_argument(
+        "--H",
+        type=int,
+        default=512,
+        help="image height, in pixel space",
+    )
+    parser.add_argument(
+        "--W",
+        type=int,
+        default=512,
+        help="image width, in pixel space",
+    )
+    parser.add_argument(
+        "--C",
+        type=int,
+        default=4,
+        help="latent channels",
+    )
+    parser.add_argument(
+        "--f",
+        type=int,
+        default=8,
+        help="downsampling factor",
+    )
+    parser.add_argument(
+        "--n_samples",
+        type=int,
+        default=3,
+        help="how many samples to produce for each given prompt. A.k.a. batch size",
+    )
+    parser.add_argument(
+        "--n_rows",
+        type=int,
+        default=0,
+        help="rows in the grid (default: n_samples)",
+    )
+    parser.add_argument(
+        "--scale",
+        type=float,
+        default=7.5,
+        help="unconditional guidance scale: eps = eps(x, empty) + scale * (eps(x, cond) - eps(x, empty))",
+    )
+    parser.add_argument(
+        "--from-file",
+        type=str,
+        help="if specified, load prompts from this file",
+    )
+    parser.add_argument(
+        "--config",
+        type=str,
+        default="configs/stable-diffusion/v1-inference.yaml",
+        help="path to config which constructs model",
+    )
+    parser.add_argument(
+        "--ckpt",
+        type=str,
+        default="models/ldm/stable-diffusion-v1/model.ckpt",
+        help="path to checkpoint of model",
+    )
+    parser.add_argument(
+        "--seed",
+        type=int,
+        default=42,
+        help="the seed (for reproducible sampling)",
+    )
+    parser.add_argument(
+        "--precision",
+        type=str,
+        help="evaluate at this precision",
+        choices=["full", "autocast"],
+        default="autocast"
+    )
+    opt = parser.parse_args()
+
+    if opt.laion400m:
+        print("Falling back to LAION 400M model...")
+        opt.config = "configs/latent-diffusion/txt2img-1p4B-eval.yaml"
+        opt.ckpt = "models/ldm/text2img-large/model.ckpt"
+        opt.outdir = "outputs/txt2img-samples-laion400m"
+
+    seed_everything(opt.seed)
+
+    config = OmegaConf.load(f"{opt.config}")
+    model = load_model_from_config(config, f"{opt.ckpt}")
+
+    device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")
+    model = model.to(device)
+
+    if opt.plms:
+        sampler = PLMSSampler(model)
+    else:
+        sampler = DDIMSampler(model)
+
+    os.makedirs(opt.outdir, exist_ok=True)
+    outpath = opt.outdir
+
+    print("Creating invisible watermark encoder (see https://github.com/ShieldMnt/invisible-watermark)...")
+    wm = "StableDiffusionV1"
+    wm_encoder = WatermarkEncoder()
+    wm_encoder.set_watermark('bytes', wm.encode('utf-8'))
+
+    batch_size = opt.n_samples
+    n_rows = opt.n_rows if opt.n_rows > 0 else batch_size
+    if not opt.from_file:
+        prompt = opt.prompt
+        assert prompt is not None
+        data = [batch_size * [prompt]]
+
+    else:
+        print(f"reading prompts from {opt.from_file}")
+        with open(opt.from_file, "r") as f:
+            data = f.read().splitlines()
+            data = list(chunk(data, batch_size))
+
+    sample_path = os.path.join(outpath, "samples")
+    os.makedirs(sample_path, exist_ok=True)
+    base_count = len(os.listdir(sample_path))
+    grid_count = len(os.listdir(outpath)) - 1
+
+    start_code = None
+    if opt.fixed_code:
+        start_code = torch.randn([opt.n_samples, opt.C, opt.H // opt.f, opt.W // opt.f], device=device)
+
+    precision_scope = autocast if opt.precision=="autocast" else nullcontext
+    with torch.no_grad():
+        with precision_scope("cuda"):
+            with model.ema_scope():
+                tic = time.time()
+                all_samples = list()
+                for n in trange(opt.n_iter, desc="Sampling"):
+                    for prompts in tqdm(data, desc="data"):
+                        uc = None
+                        if opt.scale != 1.0:
+                            uc = model.get_learned_conditioning(batch_size * [""])
+                        if isinstance(prompts, tuple):
+                            prompts = list(prompts)
+                        c = model.get_learned_conditioning(prompts)
+                        shape = [opt.C, opt.H // opt.f, opt.W // opt.f]
+                        samples_ddim, _ = sampler.sample(S=opt.ddim_steps,
+                                                         conditioning=c,
+                                                         batch_size=opt.n_samples,
+                                                         shape=shape,
+                                                         verbose=False,
+                                                         unconditional_guidance_scale=opt.scale,
+                                                         unconditional_conditioning=uc,
+                                                         eta=opt.ddim_eta,
+                                                         x_T=start_code)
+
+                        x_samples_ddim = model.decode_first_stage(samples_ddim)
+                        x_samples_ddim = torch.clamp((x_samples_ddim + 1.0) / 2.0, min=0.0, max=1.0)
+                        x_samples_ddim = x_samples_ddim.cpu().permute(0, 2, 3, 1).numpy()
+
+                        x_checked_image, has_nsfw_concept = check_safety(x_samples_ddim)
+
+                        x_checked_image_torch = torch.from_numpy(x_checked_image).permute(0, 3, 1, 2)
+
+                        if not opt.skip_save:
+                            for x_sample in x_checked_image_torch:
+                                x_sample = 255. * rearrange(x_sample.cpu().numpy(), 'c h w -> h w c')
+                                img = Image.fromarray(x_sample.astype(np.uint8))
+                                img = put_watermark(img, wm_encoder)
+                                img.save(os.path.join(sample_path, f"{base_count:05}.png"))
+                                base_count += 1
+
+                        if not opt.skip_grid:
+                            all_samples.append(x_checked_image_torch)
+
+                if not opt.skip_grid:
+                    # additionally, save as grid
+                    grid = torch.stack(all_samples, 0)
+                    grid = rearrange(grid, 'n b c h w -> (n b) c h w')
+                    grid = make_grid(grid, nrow=n_rows)
+
+                    # to image
+                    grid = 255. * rearrange(grid, 'c h w -> h w c').cpu().numpy()
+                    img = Image.fromarray(grid.astype(np.uint8))
+                    img = put_watermark(img, wm_encoder)
+                    img.save(os.path.join(outpath, f'grid-{grid_count:04}.png'))
+                    grid_count += 1
+
+                toc = time.time()
+
+    print(f"Your samples are ready and waiting for you here: \n{outpath} \n"
+          f" \nEnjoy.")
+
+
+if __name__ == "__main__":
+    main()

examples/tutorial/stable_diffusion/setup.py

+from setuptools import setup, find_packages
+
+setup(
+    name='latent-diffusion',
+    version='0.0.1',
+    description='',
+    packages=find_packages(),
+    install_requires=[
+        'torch',
+        'numpy',
+        'tqdm',
+    ],
+)
\ No newline at end of file