mayanksingal/super-slowmo

An attempt at a PyTorch implimentation of "Super SloMo: High Quality Estimation of Multiple Intermediate Frames for Video Interpolation"

super-slowmo's Issues

Can you share any downloadable database link? Thank you very much.

How much psnr and ssim do you get?

Hi, I want to know how much psnr and ssim you can get on Adobe-240 dataset after training? I have an implementation of tensorflow and the psnr is about 26 after training about 200 epoches, can't get 30+ psnr as the paper said.

Network not converge ......

Thank you for sharing your implementation.
I test the code on Adobe 240-fps data with pytorch 0.4.1. It seems that network can not converge ......
Sorry, I'm a beginner, after checking the code, I could not find the bug causing this (maybe in dataloader.populateTrainList2 12 should be replaced by 10? But this also not help.)
If it is possible, could you give me some hints why network can not converge?

Training time

Thanks for sharing
I have the same question ,the network can not converge. I have tested the code on Adobe 240-fps data about 4days. I change the batch_size = 4 ,but it also can not converge after 3days. Could you tell me the training time and the loss of your network.
Thanks a lot !

Test time

hi, how much fps can be achieved when test?

I want to know how your code differs from the original version of the pytorch code.

As far as I know you replaced resnet with vgg. in addition,
your Overall Loss：
Loss = 0.8loss_reconstruction + 0.005loss_perceptual + 0.4*loss_warping + loss_smooth
the original version of the pytorch code：
loss = 204 * recnLoss + 102 * warpLoss + 0.005 * prcpLoss + loss_smooth