search

The-Learning-And-Vision-Atelier-LAVA / SOF-VSR

[ACCV 2018 & TIP 2020] Deep Video Super-resolution using HR Optical Flow Estimation
169 stars 43 forks source link

About the draft cube (input to SRnet) #7

Closed musicrainie closed 5 years ago

musicrainie commented 5 years ago

According to the paper, the draft cube should be HxWx(2ss+1) instead of HxWx(2ss+3) in the code, which concatenates not just the central LR frame and can be fixed by changing the following line https://github.com/LongguangWang/SOF-VSR/blob/420390814ec5d0c3af6678b0773e87ca13279853/modules.py#L172 to "draft_cube = torch.unsqueeze(x[:, 1, :, :], dim=1)". Does it matter to the performance?

LongguangWang commented 5 years ago

Thank you for your attention to our work. As mentioned in README.md, we made some modifications to the original code including this one. We made this modification because we found that adding these two more channels leads to a faster convergence in our experiments. However, this modification does not matter to the performance. In our experiments, the performance is still comparable to the original one without significant improvement.

musicrainie commented 5 years ago

Ok, thanks, for your great work and nice code.