About the evaluation

[CrissyHoo]

I used the official NIQE code to evaluate the demo_000 and the result, got a unexpected result, as the niqe value of the raw video is 3.9829 while the sr video is 4.3407. I just input every frame and calculate the average value.
I don't know where is wrong, as this result is totally opposite towards that in paper.

[ckkelvinchan]

Different resolutions could lead to different NIQE. Therefore we upsample the input to maintain the same resolution during comparison. You can apply 4x upsampling first and test the NIQE.

[CrissyHoo]

Thanks for the fast response! but there are lots of different upsampling ways, could you tell me which way is used in this paper?

[ckkelvinchan]

I used the default imresize in MATLAB. That means bicubic downsampling.

[CrissyHoo]

I used the default imresize in MATLAB. That means bicubic downsampling.

hello～I have another question during testing on videolq dataset. Could you tell me the devices(how many and which type of GPUs are used when testing)? Cause I noticed that the vid4 dataset in videolq is usually used as a high quality files in most sr methods but in this paper this is regarded as the low quality one. This will lead to an increasing computation when testing.

[ckkelvinchan]

I use one V100 GPU (32GB) for test.

[CrissyHoo]

I use one V100 GPU (32GB) for test.

I noticed that you use a parameter "max_seq_len" when testing, could you tell me what's your choice? Cause I just input the whole frames of one video and then i got the error cuda out of memory with one v100 gpu. Is that the reason that you set this hyper-param?

[ckkelvinchan]

The results provided in the paper use the entire sequence. The max_seq_len is used for sequences that are much longer than 100 frames.