deepseek-ai / DreamCraft3D

[ICLR 2024] Official implementation of DreamCraft3D: Hierarchical 3D Generation with Bootstrapped Diffusion Prior
https://mrtornado24.github.io/DreamCraft3D/
MIT License
1.88k stars 81 forks source link

Where is the difference between NeRF & NeuS & Geometry stages? #59

Open yejr0229 opened 1 month ago

yejr0229 commented 1 month ago

Thanks for your great work! When reading your code, I found you divide the training into two main parts: "guidance" and "ref", which corresponding to using image generated from diffusion and using input image to supervise. The NeRF stage and NeuS stage take turns using both, Geometry stage directly using both. So I'd like to know what's the biggest difference between the NeRF stage & NeuS stage & Geometry stage?

yejr0229 commented 1 month ago
9c888773503d0f5bad903253f30481a 8fd1204e94a73b028e67cd99b02e310

Especially between NeRF and Neus, in my opinion they are a same stage, and the results of NeRF are even better than Neus.

yejr0229 commented 1 month ago

And I also found the path of "pretrained_model_name_or_path" and "pretrained_model_name_or_path_lora" in the yaml file are the same, so the lora model is not used? I didn't find any training command about dreambooth as well, when will you update it? Thanks again!