search

scenediffuser / Scene-Diffuser

Official implementation of CVPR23 paper "Diffusion-based Generation, Optimization, and Planning in 3D Scenes"
https://scenediffuser.github.io/
MIT License
338 stars 23 forks source link

Questions about the motion generation. #18

Open xjchao7 opened 2 months ago

xjchao7 commented 2 months ago

Thank you for your perfect work. I have questions about the motion generation. Can I specify starting and ending human pose to generate intermediate motion? Do I need to retrain or do I just use the trained model？

Silverster98 commented 2 months ago

Our motion generation supports two settings: (1) without using the starting position as a condition, and (2) using the starting position as a condition. If you want to specify both the starting and ending human poses, it is recommended to retrain the model by specifying both start and end as conditions.

Alternatively, you can use the diffusion inpainting capability with our pre-trained model to achieve this goal, although the results may be uncertain.