import pprint import numpy as np import torch import torchvision.transforms as T from PIL import Image from torchvision.transforms.functional import InterpolationMode from transformers import AutoModel, AutoTokenizer IMAGENET_MEAN = (0.485, 0.456, 0.406) IMAGENET_STD = (0.229, 0.224, 0.225) def build_transform(input_size): MEAN, STD = IMAGENET_MEAN, IMAGENET_STD transform = T.Compose([ T.Lambda(lambda img: img.convert('RGB') if img.mode != 'RGB' else img), T.Resize((input_size, input_size), interpolation=InterpolationMode.BICUBIC), T.ToTensor(), T.Normalize(mean=MEAN, std=STD) ]) return transform def find_closest_aspect_ratio(aspect_ratio, target_ratios, width, height, image_size): best_ratio_diff = float('inf') best_ratio = (1, 1) area = width * height for ratio in target_ratios: target_aspect_ratio = ratio[0] / ratio[1] ratio_diff = abs(aspect_ratio - target_aspect_ratio) if ratio_diff < best_ratio_diff: best_ratio_diff = ratio_diff best_ratio = ratio elif ratio_diff == best_ratio_diff: if area > 0.5 * image_size * image_size * ratio[0] * ratio[1]: best_ratio = ratio return best_ratio def dynamic_preprocess(image, min_num=1, max_num=12, image_size=448, use_thumbnail=False): orig_width, orig_height = image.size aspect_ratio = orig_width / orig_height # calculate the existing image aspect ratio target_ratios = set( (i, j) for n in range(min_num, max_num + 1) for i in range(1, n + 1) for j in range(1, n + 1) if i * j <= max_num and i * j >= min_num) target_ratios = sorted(target_ratios, key=lambda x: x[0] * x[1]) # find the closest aspect ratio to the target target_aspect_ratio = find_closest_aspect_ratio( aspect_ratio, target_ratios, orig_width, orig_height, image_size) # calculate the target width and height target_width = image_size * target_aspect_ratio[0] target_height = image_size * target_aspect_ratio[1] blocks = target_aspect_ratio[0] * target_aspect_ratio[1] # resize the image resized_img = image.resize((target_width, target_height)) processed_images = [] for i in range(blocks): box = ( (i % (target_width // image_size)) * image_size, (i // (target_width // image_size)) * image_size, ((i % (target_width // image_size)) + 1) * image_size, ((i // (target_width // image_size)) + 1) * image_size ) # split the image split_img = resized_img.crop(box) processed_images.append(split_img) assert len(processed_images) == blocks if use_thumbnail and len(processed_images) != 1: thumbnail_img = image.resize((image_size, image_size)) processed_images.append(thumbnail_img) return processed_images, target_aspect_ratio def dynamic_preprocess2(image, min_num=1, max_num=12, prior_aspect_ratio=None, image_size=448, use_thumbnail=False): orig_width, orig_height = image.size aspect_ratio = orig_width / orig_height # calculate the existing image aspect ratio target_ratios = set( (i, j) for n in range(min_num, max_num + 1) for i in range(1, n + 1) for j in range(1, n + 1) if i * j <= max_num and i * j >= min_num) target_ratios = sorted(target_ratios, key=lambda x: x[0] * x[1]) new_target_ratios = [] for i in target_ratios: if prior_aspect_ratio[0] % i[0] or prior_aspect_ratio[1] % i[1]: new_target_ratios.append(i) else: continue # find the closest aspect ratio to the target target_aspect_ratio = find_closest_aspect_ratio( aspect_ratio, new_target_ratios, orig_width, orig_height, image_size) # calculate the target width and height target_width = image_size * target_aspect_ratio[0] target_height = image_size * target_aspect_ratio[1] blocks = target_aspect_ratio[0] * target_aspect_ratio[1] # resize the image resized_img = image.resize((target_width, target_height)) processed_images = [] for i in range(blocks): box = ( (i % (target_width // image_size)) * image_size, (i // (target_width // image_size)) * image_size, ((i % (target_width // image_size)) + 1) * image_size, ((i // (target_width // image_size)) + 1) * image_size ) # split the image split_img = resized_img.crop(box) processed_images.append(split_img) assert len(processed_images) == blocks if use_thumbnail and len(processed_images) != 1: thumbnail_img = image.resize((image_size, image_size)) processed_images.append(thumbnail_img) return processed_images def load_image(image, input_size=448, min_num=1, max_num=12): image = image.convert('RGB') transform = build_transform(input_size=input_size) images, target_aspect_ratio = dynamic_preprocess(image, image_size=input_size, use_thumbnail=True, min_num=min_num, max_num=max_num) pixel_values = [transform(image) for image in images] pixel_values = torch.stack(pixel_values) return pixel_values, target_aspect_ratio def load_image2(image, input_size=448, min_num=1, max_num=12, target_aspect_ratio=None): image = image.convert('RGB') transform = build_transform(input_size=input_size) images = dynamic_preprocess2(image, image_size=input_size, use_thumbnail=True, min_num=min_num, max_num=max_num, prior_aspect_ratio=target_aspect_ratio) pixel_values = [transform(image) for image in images] pixel_values = torch.stack(pixel_values) return pixel_values import warnings from .base import BaseModel from ..dataset import DATASET_TYPE class MiniMonkey(BaseModel): INSTALL_REQ = False INTERLEAVE = False def __init__(self, model_path='mx262/MiniMonkey', **kwargs): assert model_path is not None self.model_path = model_path self.model_type = torch.bfloat16 self.model = AutoModel.from_pretrained( self.model_path, low_cpu_mem_usage=True, trust_remote_code=True).eval().to(self.model_type).cuda() self.tokenizer = AutoTokenizer.from_pretrained(self.model_path, trust_remote_code=True, use_fast=False) self.kwargs = kwargs warnings.warn(f'Following kwargs received: {self.kwargs}, will use as generation config. ') torch.cuda.empty_cache() def generate_inner(self, message, dataset=None): prompt, image_path = self.message_to_promptimg(message, dataset=dataset) if dataset is None: return self.generate_vanilla(image_path, prompt) assert isinstance(dataset, str) if DATASET_TYPE(dataset) == 'MCQ' or DATASET_TYPE(dataset) == 'Y/N' or dataset == 'HallusionBench': return self.generate_multichoice(image_path, prompt) else: return self.generate_vanilla(image_path, prompt) def generate_vanilla(self, image_path, prompt): image = Image.open(image_path).convert('RGB') pixel_values, target_aspect_ratio = load_image(image, min_num=4, max_num=12) pixel_values = pixel_values.cuda().to(self.model_type) pixel_values2 = load_image2(image, min_num=3, max_num=7, target_aspect_ratio=target_aspect_ratio) pixel_values2 = pixel_values2.cuda().to(self.model_type) pixel_values = torch.cat([pixel_values2[:-1], pixel_values[:-1], pixel_values2[-1:]], 0) generation_config = dict(do_sample=False, max_new_tokens=512) response, history = self.model.chat(self.tokenizer, pixel_values, target_aspect_ratio, prompt, generation_config, history=None, return_history=True) return response def generate_multichoice(self, image_path, prompt): return self.generate_vanilla(image_path, prompt)