search

showlab / MotionDirector

[ECCV 2024 Oral] MotionDirector: Motion Customization of Text-to-Video Diffusion Models.
https://showlab.github.io/MotionDirector/
Apache License 2.0
850 stars 54 forks source link

cache_latent is doing for what #46

Open JustinKai0527 opened 3 weeks ago

JustinKai0527 commented 3 weeks ago

I found I use the config_multi_image.yaml they got multple cache what is this for why have many cache? is this for inference? image but in the above image is just load the lora weight for subject spatial and motion temporal

ruizhaocv commented 1 week ago

These caches are used in training to avoid encoding training data for multiple times, and they are not used during inference.