Evaluation Results for gopro dataset

[VongolaWu]

Hi, When I use your provided checkpoint for gopro dataset, the evaluate result seems very low. I set the resume_epoch to 599 (epochs=600) and resume_path to the provided checkpoint (checkpoints/IFRNet_GoPro.pth), also comments all the training parts so that it will directly go to the evaluation part. The results are: Could you give me some possible reasons so that I can debug?

[ltkong218]

For your problems, I have following suggestions:

We have tested our pre-trained checkpoints of IFRNet for both 2x and 8x frame interpolation. If your PSNR is only around 15~16 dB, I think your experimental environment is incorrectly configured. Please check whether your PyTorch version and backward warping operation meet the requirement. If your environment is correctly configured, you should reproduce our 2x and 8x frame interpolation demos in our GitHub repository. Please also follow the training and test datasets split in our code.

Thanks.

[VongolaWu]

Thank you for your reply. Could you share the environment list? So that I can use exact packages.

[ltkong218]

In fact, I have developed and deployed this repository on different environment including PyTorch 1.3.0 and PyTorch 1.9.1. Since the IFRNet is concise and does not depend on complex modules, most PyTorch environments can run IFRNet. Thanks.