Songwei Ge

Thanks for the detailed comments! These are definitely useful!!

Thanks for the suggestion! I will look into that if I have any time in the future!!

Thank you for providing the detailed log and trying to solve it with the update. I think there is a missing package in the environment. I have updated the `install.py`....

I have the same question here. Seems that there is a discrepancy in the preprocessing code between the [training](https://github.com/OpenGVLab/VideoMAEv2/blob/bddefef66ea189a7d83f56bf1769b25594bedfdc/dataset/pretrain_datasets.py#L36) and [inference](https://github.com/OpenGVLab/VideoMAEv2/blob/bddefef66ea189a7d83f56bf1769b25594bedfdc/extract_tad_feature.py#L105).

Hi @Alienge, how many samples did you use to compute the FVD? It seems that there is a numerical error where it could happen in the covariance matric inversion step...

Thank you for digging into this and sharing what you found!! I suspect this was because of a loose check in real->complex conversion for Schur decomposition, which only happens in...

I have the same issue and was only able to do ``pip install`` when the versions are consistent with the python in base env.

At this moment, both [I3D](https://github.com/songweige/content-debiased-fvd/blob/118f4ac7ce07c80eff44118a47016257438fb01d/cdfvd/third_party/i3d/utils.py#L20) and [VideoMAEv2](https://github.com/songweige/content-debiased-fvd/blob/118f4ac7ce07c80eff44118a47016257438fb01d/cdfvd/third_party/VideoMAEv2/utils.py#L42) preprocess videos into 224 resolution. In practice, I see people use the same metric computation for videos with higher resolutions without resizing. Let...