atelili/2bivqa

2BiVQA is a no-reference deep learning based video quality assessment metric.

2bivqa's Issues

How to generate konvid.csv for our case?

I run the feature extraction script with following command, but don't know how to generate that konvid.csv for further processing. Please help me out.

!python3 extract_features.py -v /content/drive/MyDrive/videos -f konvid.csv -o 0.2 -fl 0

spatial_train

Hello, I want to train my own spatial pooling model, but I don't know how to extract image features!，The following figure shows that I extract features from the video, and then_ Results of spatial_train.py training.It can be seen that my feature extraction is wrong, so I want to extract features from the image. How can I do itimage

model

Hello, I want to ask about the test. I used the spatial_model.h5 model I trained myself, but it can't run. Could you help me see what the problem is

test_model

Hello, I want to ask test_ How to make the test file in the model file. I think the shape of x in your Konvid dataset is (7200，51200)

extract_feaures

Hello, I want to ask if you want to extract and save video frames locally in advance. I tried to extract and save the video frame to the local, but the following error occurred when running
It should be that I didn't read the image, so first ask me what I should do to extract features. This is the video frame I extracted
The underlined part is the name of the video

test

Hello, I want to ask how to generate the characteristic file of the test set corresponding to each data set, such as the one you provided

how to change the input size in pretrained model?

hi~
I would like to do some experiments with your code. I want to change the input size of (30, 25, 2048), but in the pre-trained model 'res-bi-sp_koniq.h5', this input seems to be fixed, how can I get my own pre-trained model with arbitrary size?

issue

What causes the model training not to stop

spatial_train

Hello, I want to train my own spatial pooling model, but I don't know how to extract image features!，The following figure shows that I extract features from the video, and then_ Results of spatial_train.py training.It can be seen that my feature extraction is wrong, so I want to extract features from the image. How can I do itimage

A great work !

Thanks for your great work ! Looking forward to the article for learning more details.

about the paper

I am interested in your work and would like to read the paper for more information. When will the paper be open sourced? Thanks~

pre-trained models on Youtobe-UGC

Hello, I would like to inquire if it is possible to provide the pre-trained models on UGC.

What was the best score of the size nb_frames and overlap?

Hi, may you tell me the number of overlap and nb_frames?

Extract features

I have set the flag to 1, but there is still a problem. I extracted the video in KoNViD into an image and saved it. Excuse me, this is how it works