|
|
import torch |
|
|
from tqdm import tqdm |
|
|
from torchvision import transforms |
|
|
from pyiqa.archs.musiq_arch import MUSIQ |
|
|
from vbench.utils import load_video, load_dimension_info, CACHE_DIR |
|
|
|
|
|
def transform(images, preprocess_mode='shorter'): |
|
|
if preprocess_mode.startswith('shorter'): |
|
|
_, _, h, w = images.size() |
|
|
if min(h,w) > 512: |
|
|
scale = 512./min(h,w) |
|
|
images = transforms.Resize(size=( int(scale * h), int(scale * w) ))(images) |
|
|
if preprocess_mode == 'shorter_centercrop': |
|
|
images = transforms.CenterCrop(512)(images) |
|
|
|
|
|
elif preprocess_mode == 'longer': |
|
|
_, _, h, w = images.size() |
|
|
if max(h,w) > 512: |
|
|
scale = 512./max(h,w) |
|
|
images = transforms.Resize(size=( int(scale * h), int(scale * w) ))(images) |
|
|
|
|
|
elif preprocess_mode == 'None': |
|
|
return images / 255. |
|
|
|
|
|
else: |
|
|
raise ValueError("Please recheck imaging_quality_mode") |
|
|
return images / 255. |
|
|
|
|
|
def technical_quality(model, video_pairs, device, **kwargs): |
|
|
if 'imaging_quality_preprocessing_mode' not in kwargs: |
|
|
preprocess_mode = 'longer' |
|
|
else: |
|
|
preprocess_mode = kwargs['imaging_quality_preprocessing_mode'] |
|
|
video_results = [] |
|
|
|
|
|
for info in tqdm(video_pairs): |
|
|
query = info['prompt'] |
|
|
video_path = info['content_path'] |
|
|
|
|
|
images = load_video(video_path) |
|
|
images = transform(images, preprocess_mode) |
|
|
acc_score_video = 0. |
|
|
for i in range(len(images)): |
|
|
frame = images[i].unsqueeze(0).to(device) |
|
|
score = model(frame) |
|
|
acc_score_video += float(score) |
|
|
|
|
|
video_result = acc_score_video/len(images) |
|
|
video_results.append({'prompt':query, 'video_path': video_path, 'video_results': video_result/100.}) |
|
|
|
|
|
|
|
|
average_score = sum([o['video_results'] for o in video_results]) / len(video_results) |
|
|
|
|
|
|
|
|
return { |
|
|
"score":[average_score, video_results] |
|
|
} |
|
|
|
|
|
|
|
|
def compute_imaging_quality(video_pairs): |
|
|
device = torch.device("cuda") |
|
|
model_path = f'{CACHE_DIR}/pyiqa_model/musiq_spaq_ckpt-358bb6af.pth' |
|
|
kwargs = { |
|
|
'imaging_quality_preprocessing_mode' : 'longer' |
|
|
} |
|
|
|
|
|
model = MUSIQ(pretrained_model_path=model_path) |
|
|
model.to(device) |
|
|
model.training = False |
|
|
|
|
|
results = technical_quality(model, video_pairs, device, **kwargs) |
|
|
return results |
|
|
|
