search

nihaomiao / CVPR23_LFDM

The pytorch implementation of our CVPR 2023 paper "Conditional Image-to-Video Generation with Latent Flow Diffusion Models"
BSD 2-Clause "Simplified" License
432 stars 43 forks source link

Evaluation #27

Closed SPengLiang closed 10 months ago

SPengLiang commented 11 months ago

Hi, thanks for your great work! How can I evaluate the test results? I did not find scripts/codes for quantitative evaluation for mhad dataset.

nihaomiao commented 11 months ago

Hi, @SPengLiang, thanks for your interest in our work! I use the FVD code provided here to evaluate our models. You need to generate a number of fake videos first and then compare their distribution with that of real videos. The subject FVD and class FVD are just simple variants of the FVD score. You may check our paper for more details.

SPengLiang commented 11 months ago

@nihaomiao Thanks a lot!

SPengLiang commented 10 months ago

@nihaomiao Using this function (fvd = compute_our_fvd(model.real_out_vid.data.cpu().numpy().copy(), real_vids.data.cpu().numpy().copy())), I got an error: "RuntimeError: Given groups=1, weight of size [64, 3, 7, 7, 7], expected input[10, 128, 9, 45, 133] to have 3 channels, but got 128 channels instead ".
The shapes of input videos are (10, 3, 40, 128, 128) and (10, 3, 40, 128, 128). Do I miss something?