| import io | |
| import os | |
| import cv2 | |
| import json | |
| import numpy as np | |
| from PIL import Image | |
| from tqdm import tqdm | |
| import torch | |
| import torch.nn as nn | |
| import torch.nn.functional as F | |
| import torchvision.transforms as transforms | |
| from vbench.utils import load_video, load_dimension_info, dino_transform, dino_transform_Image | |
| import logging | |
| logging.basicConfig(level = logging.INFO,format = '%(asctime)s - %(name)s - %(levelname)s - %(message)s') | |
| logger = logging.getLogger(__name__) | |
| def subject_consistency(model, video_pairs, device): | |
| sim = 0.0 | |
| cnt = 0 | |
| video_results = [] | |
| image_transform = dino_transform(224) | |
| for info in tqdm(video_pairs): | |
| query = info['prompt'] | |
| video_path = info['content_path'] | |
| video_sim = 0.0 | |
| images = load_video(video_path) | |
| images = image_transform(images) | |
| for i in range(len(images)): | |
| with torch.no_grad(): | |
| image = images[i].unsqueeze(0) | |
| image = image.to(device) | |
| image_features = model(image) | |
| image_features = F.normalize(image_features, dim=-1, p=2) | |
| if i == 0: | |
| first_image_features = image_features | |
| else: | |
| sim_pre = max(0.0, F.cosine_similarity(former_image_features, image_features).item()) | |
| sim_fir = max(0.0, F.cosine_similarity(first_image_features, image_features).item()) | |
| cur_sim = (sim_pre + sim_fir) / 2 | |
| video_sim += cur_sim | |
| cnt += 1 | |
| former_image_features = image_features | |
| sim_per_images = video_sim / (len(images) - 1) | |
| sim += video_sim | |
| video_results.append({'prompt':query, 'video_path': video_path, 'video_results': sim_per_images}) | |
| sim_per_frame = sim / cnt | |
| return { | |
| "score":[sim_per_frame, video_results] | |
| } | |
| def compute_subject_consistency(video_pairs): | |
| device = torch.device("cuda") | |
| submodules_list = { | |
| 'repo_or_dir':'facebookresearch/dino:main', | |
| 'source':'github', | |
| 'model': 'dino_vitb16', | |
| } | |
| dino_model = torch.hub.load(**submodules_list).to(device) | |
| logger.info("Initialize DINO success") | |
| results = subject_consistency(dino_model, video_pairs, device) | |
| return results | |