search

Vchitect / Latte

Latte: Latent Diffusion Transformer for Video Generation.
Apache License 2.0
1.44k stars 147 forks source link

how to get 2048 videos for computing FVD? #43

Open hdjsjyl opened 4 months ago

hdjsjyl commented 4 months ago

Hi author, Thanks for your great work. I would like to know how to get 2048 videos for computing FVD? Thanks

maxin-cn commented 4 months ago

Hi author, Thanks for your great work. I would like to know how to get 2048 videos for computing FVD? Thanks

Just run bash sample/ffs_ddp.sh or bash sample/ffs_ddp.sh.

hdjsjyl commented 4 months ago

Hi @maxin-cn , Thanks for your reply. My questions is how to get corresponding 2048 ground truth data, not generated data. In this paper, it says 2048 real videos with 16 frames. I am not sure about how to get it. Any help is appreciated, thanks

maxin-cn commented 4 months ago

Hi @maxin-cn , Thanks for your reply. My questions is how to get corresponding 2048 ground truth data, not generated data. In this paper, it says 2048 real videos with 16 frames. I am not sure about how to get it. Any help is appreciated, thanks

You can randomly sample 2048 videos from the real dataset or also use the entire real videos to calculate FVD.