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

'jitter' effect #45

Closed h-bouzid closed 1 month ago

h-bouzid commented 2 months ago

Hi dear @nihaomiao

Congratulations on your very interesting work! I'm curious about the role of data jittering. If it's to prevent model overfitting, why was it implemented only in the DM and not in the LFAE? and have you tested running the DM without jittering the data?

Thank you in advance 😁

nihaomiao commented 2 months ago

Hi, @h-bouzid, thanks a lot for your interest in our work! To be honest, I have not noticed this point before. But LFAE is initialized with a pretrained model. It would be OK without using "jitter". I have not tested training the DM without using "jitter". In general, data augmentation should not have negative effects.

h-bouzid commented 1 month ago

Thank you for your answer @nihaomiao.