WU-CVGL / MVControl

[3DV-2025] Official implementation of "Controllable Text-to-3D Generation via Surface-Aligned Gaussian Splatting"
https://lizhiqi49.github.io/MVControl/
MIT License
191 stars 6 forks source link

Questions about using multi-view images as MVControlNet's condition. #11

Open snowflakewang opened 3 months ago

snowflakewang commented 3 months ago

Hello, thank you for your work on adding a ControlNet on MVDream. I am curious about using multi-view (4-view for MVDream) images instead of only one image as input for MVControlNet, since MVDream can handle multi-view images during the denoising process and ControlNet is a trainable copy of the architecture of MVDream. If I want to fine-tune (MVDream + MVControlNet) to achieve multi-view X to multi-view Y, is the MVControl compatible to it?

Thank you!