zkf1997 / DIMOS

code release of paper "DIMOS: Synthesizing Diverse Human Motions in 3D Indoor Scenes"
https://zkf1997.github.io/DIMOS/
Apache License 2.0
79 stars 3 forks source link

About training cost #5

Open wangzhidi opened 8 months ago

wangzhidi commented 8 months ago

I am attempting to reproduce your work, and I would like to know about the training resources utilized for the CVAE and reinforcement learning components in your work, such as the duration of training and the graphics card used.

zkf1997 commented 8 months ago

I trained with a single 4090. The marker predictor training time depends on the amount of training data and can be finished in several hours. The policy training cost 10-20 hours each according to my experience.

wangzhidi commented 8 months ago

Thanks for answering! This is excellent work, I am following you now.