junpan19 / rrn Goto Github PK

Thanks

Hi, could you please release the code of your paper Multi-Target Domain Adaptation with Collaborative Consistency Learning? The link repository that you published on the paper (https://github.com/junpan19/MTDA) does not work.

Hello,
MSU Video Group has recently launched Video Super Resolution Benchmark and evaluated this algorithm.

RRN-10L takes 6th place by subjective score, 13th place by PSNR, and 12th by our metric ERQAv1.0. RRN-5L takes 9th place by subjective score, 14th place by PSNR, and 13th by our metric ERQAv1.0. You can see the results here.

If you have any other VSR method you want to see in our benchmark, we kindly invite you to participate.
You can submit it for the benchmark, following the submission steps.

if self.scale == 4:
     target = np.lib.pad(target, pad_width=((0,0), (2*self.scale,2*self.scale), (2*self.scale,2*self.scale), (0,0)), mode='reflect')

1. when I delete the code. it means without using reflect pad. Then the result decline to 24.73 in Vid4
   use pretrain model RRN-10L.pth

==> Average PSNR = 24.731251
==> Average SSIM = 0.816482

2. where is the code [64x64 patch of LR]

Hi! I've been working with these scripts! It's wonderful work! But I do have a question. I can't seem to get the scale function to do anything. Am I do something wrong, or is it non-functional?

It seems like the operation of 256x256 random cropping on Vimeo90K HR images is conducted before the model training process, instead of using an online fashion. The codes do not give descriptions on how this is done.

Could you explain the details of data pre-processing about (1) where to crop 256*256 (are image corners avoided?) ; (2) how many cropped videos are generated from the original videos?

I am now generating LR-HR pairs on-the-fly, so each vimeo clip is fetched only once during one epoch. The PSNR results on Vid4 Y channel (25.7 reproduced) are far worse than reported (27.69) after 70 epochs. I wonder whether this is due to inadequate training epochs since only 1 cropped clip for each HR clip is used.

If there are some tricks that the code not contain?

after training i found SSIM and PSNR is lower than paper，i‘ve trained 180 epoch, i want to know what's the reason?

if it's because i use the wrong way to crop vimeo90k, i use PIL : img_src.crop((0,0,256,256))