search

ashawkey / stable-dreamfusion

Text-to-3D & Image-to-3D & Mesh Exportation with NeRF + Diffusion.
Apache License 2.0
7.99k stars 710 forks source link

dreambooth3d #183

Open wlupc opened 1 year ago

wlupc commented 1 year ago

The stable-dreamfusion is a great work. I wonder to know the differences between Dreamfusion and dreambooth3d. Looking forward to receiving your response.

MathieuTuli commented 1 year ago

scanning the dreambooth3d paper, there's two main differences:

  1. Using DreamBooth rather than Stable Diffusion (DreamFusion paper actually uses Imagen). This is already a feature in this repo I believe, where you can swap out any diffusion model you want.
  2. Multi-staged training. They do a two-stage training where they initially train a nerf for a secondary data-augmented fine-tuning stage, where in each stage a varied-training version of the dreambooth model is used. I didn't read the whole thing but it seems they do this for multi-view consistency.
zeng-yifei commented 1 year ago

Yep, dreambooth3d is actually a dreamfusion pro+. It mainly focus on how to finetune a dreamfusion pipeline.