transforms.py

from __future__ import division
import torch
import math
import random
from PIL import Image, ImageOps
try:
    import accimage
except ImportError:
    accimage = None
import numpy as np
import numbers
import types
import collections


class Compose(object):
    """Composes several transforms together.

    Args:
        transforms (List[Transform]): list of transforms to compose.

    Example:
        >>> transforms.Compose([
        >>>     transforms.CenterCrop(10),
        >>>     transforms.ToTensor(),
        >>> ])
    """

    def __init__(self, transforms):
        self.transforms = transforms

    def __call__(self, img):
        for t in self.transforms:
            img = t(img)
        return img


class ToTensor(object):
    """Converts a PIL.Image or numpy.ndarray (H x W x C) in the range
    [0, 255] to a torch.FloatTensor of shape (C x H x W) in the range [0.0, 1.0].
    """

    def __call__(self, pic):
        if isinstance(pic, np.ndarray):
            # handle numpy array
            img = torch.from_numpy(pic.transpose((2, 0, 1)))
            # backward compatibility
            return img.float().div(255)

        if accimage is not None and isinstance(pic, accimage.Image):
            nppic = np.zeros([pic.channels, pic.height, pic.width], dtype=np.float32)
            pic.copyto(nppic)
            return torch.from_numpy(nppic)

        # handle PIL Image
        if pic.mode == 'I':
            img = torch.from_numpy(np.array(pic, np.int32, copy=False))
        elif pic.mode == 'I;16':
            img = torch.from_numpy(np.array(pic, np.int16, copy=False))
        else:
            img = torch.ByteTensor(torch.ByteStorage.from_buffer(pic.tobytes()))
        # PIL image mode: 1, L, P, I, F, RGB, YCbCr, RGBA, CMYK
        if pic.mode == 'YCbCr':
            nchannel = 3
        elif pic.mode == 'I;16':
            nchannel = 1
        else:
            nchannel = len(pic.mode)
        img = img.view(pic.size[1], pic.size[0], nchannel)
        # put it from HWC to CHW format
        # yikes, this transpose takes 80% of the loading time/CPU
        img = img.transpose(0, 1).transpose(0, 2).contiguous()
        if isinstance(img, torch.ByteTensor):
            return img.float().div(255)
        else:
            return img


class ToPILImage(object):
    """Converts a torch.*Tensor of shape C x H x W or a numpy ndarray of shape
    H x W x C to a PIL.Image while preserving value range.
    """

    def __call__(self, pic):
        npimg = pic
        mode = None
        if isinstance(pic, torch.FloatTensor):
            pic = pic.mul(255).byte()
        if torch.is_tensor(pic):
            npimg = np.transpose(pic.numpy(), (1, 2, 0))
        assert isinstance(npimg, np.ndarray), 'pic should be Tensor or ndarray'
        if npimg.shape[2] == 1:
            npimg = npimg[:, :, 0]

            if npimg.dtype == np.uint8:
                mode = 'L'
            if npimg.dtype == np.int16:
                mode = 'I;16'
            if npimg.dtype == np.int32:
                mode = 'I'
            elif npimg.dtype == np.float32:
                mode = 'F'
        else:
            if npimg.dtype == np.uint8:
                mode = 'RGB'
        assert mode is not None, '{} is not supported'.format(npimg.dtype)
        return Image.fromarray(npimg, mode=mode)


class Normalize(object):
    """Given mean: (R, G, B) and std: (R, G, B),
    will normalize each channel of the torch.*Tensor, i.e.
    channel = (channel - mean) / std
    """

    def __init__(self, mean, std):
        self.mean = mean
        self.std = std

    def __call__(self, tensor):
        # TODO: make efficient
        for t, m, s in zip(tensor, self.mean, self.std):
            t.sub_(m).div_(s)
        return tensor


class Scale(object):
    """Rescales the input PIL.Image to the given 'size'.
    If 'size' is a 2-element tuple or list in the order of (width, height), it will be the exactly size to scale.
    If 'size' is a number, it will indicate the size of the smaller edge.
    For example, if height > width, then image will be
    rescaled to (size * height / width, size)
    size: size of the exactly size or the smaller edge
    interpolation: Default: PIL.Image.BILINEAR
    """

    def __init__(self, size, interpolation=Image.BILINEAR):
        assert isinstance(size, int) or (isinstance(size, collections.Iterable) and len(size) == 2)
        self.size = size
        self.interpolation = interpolation

    def __call__(self, img):
        if isinstance(self.size, int):
            w, h = img.size
            if (w <= h and w == self.size) or (h <= w and h == self.size):
                return img
            if w < h:
                ow = self.size
                oh = int(self.size * h / w)
                return img.resize((ow, oh), self.interpolation)
            else:
                oh = self.size
                ow = int(self.size * w / h)
                return img.resize((ow, oh), self.interpolation)
        else:
            return img.resize(self.size, self.interpolation)


class CenterCrop(object):
    """Crops the given PIL.Image at the center to have a region of
    the given size. size can be a tuple (target_height, target_width)
    or an integer, in which case the target will be of a square shape (size, size)
    """

    def __init__(self, size):
        if isinstance(size, numbers.Number):
            self.size = (int(size), int(size))
        else:
            self.size = size

    def __call__(self, img):
        w, h = img.size
        th, tw = self.size
        x1 = int(round((w - tw) / 2.))
        y1 = int(round((h - th) / 2.))
        return img.crop((x1, y1, x1 + tw, y1 + th))


class Pad(object):
    """Pads the given PIL.Image on all sides with the given "pad" value"""

    def __init__(self, padding, fill=0):
        assert isinstance(padding, numbers.Number)
        assert isinstance(fill, numbers.Number) or isinstance(fill, str) or isinstance(fill, tuple)
        self.padding = padding
        self.fill = fill

    def __call__(self, img):
        return ImageOps.expand(img, border=self.padding, fill=self.fill)


class Lambda(object):
    """Applies a lambda as a transform."""

    def __init__(self, lambd):
        assert isinstance(lambd, types.LambdaType)
        self.lambd = lambd

    def __call__(self, img):
        return self.lambd(img)


class RandomCrop(object):
    """Crops the given PIL.Image at a random location to have a region of
    the given size. size can be a tuple (target_height, target_width)
    or an integer, in which case the target will be of a square shape (size, size)
    """

    def __init__(self, size, padding=0):
        if isinstance(size, numbers.Number):
            self.size = (int(size), int(size))
        else:
            self.size = size
        self.padding = padding

    def __call__(self, img):
        if self.padding > 0:
            img = ImageOps.expand(img, border=self.padding, fill=0)

        w, h = img.size
        th, tw = self.size
        if w == tw and h == th:
            return img

        x1 = random.randint(0, w - tw)
        y1 = random.randint(0, h - th)
        return img.crop((x1, y1, x1 + tw, y1 + th))


class RandomHorizontalFlip(object):
    """Randomly horizontally flips the given PIL.Image with a probability of 0.5
    """

    def __call__(self, img):
        if random.random() < 0.5:
            return img.transpose(Image.FLIP_LEFT_RIGHT)
        return img


class RandomSizedCrop(object):
    """Random crop the given PIL.Image to a random size of (0.08 to 1.0) of the original size
    and and a random aspect ratio of 3/4 to 4/3 of the original aspect ratio
    This is popularly used to train the Inception networks
    size: size of the smaller edge
    interpolation: Default: PIL.Image.BILINEAR
    """

    def __init__(self, size, interpolation=Image.BILINEAR):
        self.size = size
        self.interpolation = interpolation

    def __call__(self, img):
        for attempt in range(10):
            area = img.size[0] * img.size[1]
            target_area = random.uniform(0.08, 1.0) * area
            aspect_ratio = random.uniform(3. / 4, 4. / 3)

            w = int(round(math.sqrt(target_area * aspect_ratio)))
            h = int(round(math.sqrt(target_area / aspect_ratio)))

            if random.random() < 0.5:
                w, h = h, w

            if w <= img.size[0] and h <= img.size[1]:
                x1 = random.randint(0, img.size[0] - w)
                y1 = random.randint(0, img.size[1] - h)

                img = img.crop((x1, y1, x1 + w, y1 + h))
                assert(img.size == (w, h))

                return img.resize((self.size, self.size), self.interpolation)

        # Fallback
        scale = Scale(self.size, interpolation=self.interpolation)
        crop = CenterCrop(self.size)
        return crop(scale(img))