zhangzjn / dtvnet Goto Github PK

When generating videos, I found out that the pretrained weights "Flow_G.pth" are not loaded. Is this an expected behavior of the model?

hello, I want to know How do you test PSNR，SSIM, could you republic the code?
when I test all the "sky-test" datasets(the length is 4864), the "PSNR"(about 23 in my test) and "SSIM"(about 0.7 in my test) are lower than paper, and I saw the file(psnr.csv), is your test datasets?
Hope to get your reply， thank you！

I've tried generating videos using the provided pretrained model, but the result is not as clean as the one provided in the sample output, and the PSNR and SSIM values are quite low (lower than the results provided by the paper).

I used the provided "DTV_G.pth" and optical flow npy files to generate the videos. I also used the parameters described in configs/sky_timelapse.json without any changes.

The video I generated is the bottom one. I've also put the sample output video for reference. (upper: sample output, bottom: my result)

Am I generating the videos wrongly by any means?

I cannot find the pre-trained models that are supposed to be at checkpoints/DTV_Sky/200708162546/results. I have tried to grep the entire repo for the files/folder and haven't found them.

Hello,
Thank you for sharing your code and dataset. I could not download your dataset because I need to have a Chinese phone number to register in the website and then download the dataset. May you please help me through another way?
Thank you,
Nuha

As described in the paper:

" We split the selected video clips into a training set (containing 1000 clips, totally 244,930 frames), a validation set (containing 100 clips, totally 23,200 frames), and a testing set (containing 67 clips, totally 17,316 frames)."

There are 67 clips for testing and 100 clips for validation. However, the downloaded meta files (test_urls.txt and val_urls.txt) show that there are 100 clips for testing and 67 for validation.

Which version is correct?