lidq92 / mdtvsfa Goto Github PK

我只想训练konvid-1k,�但是你们的网络里面有个key，导致我必须要建立一个key

@lidq92 Thanks for your sharing. I found that 'CNN_feature_extract' and 'regression_score' are carried out separately during the training. Have you considered an end-to-end network? That means training the ‘CNN_feature_extract’ network. I want to know if it is necessary to train the feature extraction part.Thank you very much!!

ERROR: Unexpected bus error encountered in worker. This might be caused by insufficient shared memory (shm).
查了我的docker容器默认shm是64M，请问应该改为多少G才不会报这个错

前辈您好，请问为什么要用方差做质量评价

@lidq92 ,Hi, thanks for your sharing，and I found that the code is running on a single gpu ，why not run on multiple GPUs? I can successfully run the CNNfeature.py on a single gpu，but when I use multiprocessing , I meet the error:

THCudaCheck FAIL file=/pytorch/aten/src/THC/THCGeneral.cpp line=50 error=3 : initialization error
Process Process-8:
Traceback (most recent call last):
File "/root/anaconda3/lib/python3.6/multiprocessing/process.py", line 258, in _bootstrap
self.run()
File "/root/anaconda3/lib/python3.6/multiprocessing/process.py", line 93, in run
self._target(*self._args, **self._kwargs)
File "/MDTVSFA-main/CNNfeatures_mp.py", line 147, in run
features = get_features(current_data['video'], frame_batch_size, model, device)
File "/MDTVSFA-main/CNNfeatures_mp.py", line 121, in get_features
output1 = torch.Tensor().to(device)
File "/root/anaconda3/lib/python3.6/site-packages/torch/cuda/init.py", line 193, in _lazy_init
torch._C._cuda_init()
RuntimeError: cuda runtime error (3) : initialization error at /pytorch/aten/src/THC/THCGeneral.cpp:50

@lidq92 Thanks for your reply. When I train the network end to end ,that means I will train the Resnet-50,overfifting will occur,SROCC is 0.1.But if I just only train the regression net, the SROCC is 0.6. What should I do?Thank you very much!!

@lidq92 Thanks for your sharing. The article says before and after frames are taken into account.（《Qality Assessment of In-the-Wild Videos》formula 7，8，9）But I find the code consider only one frame,relative_score[d][i] = self._sitp(q[i, :x_len[i].item()]).I can't find any code to store the before and after frames. Would you please explain it for me? Thank you very much!!

Hello! We have recently launched and evaluated this algorithm on the dataset of our video quality metrics benchmark. The dataset distortions refer to compression artifacts on professional and user-generated content. Method took 3th place on the global leaderboard and 1th place on the no-reference-only leaderboard in the terms of SROCC. You can see more detailed results here. If you have any other video quality metric (either full-reference or no-reference) that you want to see in our benchmark, we kindly invite you to participate. You can submit it to the benchmark, following the submission steps, described here.

请问您有KonVid-150k数据集的下载链接嘛

m = train_dataset.m
m代表什么，为什么可以train_dataset后面直接加.m

@lidq92 Thanks for your sharing. I found that :
if self.loss_type == 'mixed': loss = [loss_a(mapped_score[d], y[d]) + loss_m(relative_score[d], y[d]) + F.l1_loss(aligned_score[d], y[d]) / self.scale[d] for d in range(len(y))]
mapped_score[d]：0--1
mapped_score[d]：0--1
aligned_score[d]：1--5
y：1--5
The scale is not uniform, do you need to scale before calculating? I can't understand why the first and second terms are not divided by self.scale[d], but the third term is divided by self.scale[d]. Thank you very much!!

@lidq92 Thanks for your sharing . The paper says that comparative experiment was done using BRISQUE features. How to extract BRISQUE features, using MATLAB or Python code, could you share the code with me? Thank you very much!!

@lidq92 Thanks for your sharing.

1. I find that the "skvideo.io.vread" is used in the project. And I want to know if it extract three channels or one channel when the video format is YUV420p.
2. And I find it faster to extract video frames using opencv, why did you choose "skvideo.io.vread"? There is a little bit of difference between "skvideo.io.vread" and "opencv"(0.47704992 VS 0.47492328 ).
3. I don't know the meaning of the frame = Image.fromarray(frame), because when I remove this code ,I still get the same result.

Thank you very much!!

Hi Postdoc Li, why there is no key of ['VQA_performance'] in evaluator.state.metrics ？