Attach gaussians and refine configuration

[xodus777]

Hi @lizhiqi49,

Thank you for your excellent work. I’m running into an issue during the Gaussian Splatting refinement stage. I’m using 21.png as the static reconstruction source for Aurorus—while the mesh itself reconstructs correctly, the refinement step doesn’t seem to execute.

Could you share your sugar_static_refine.yaml configuration? I suspect I might have misconfigured the system.guidance.pretrained_model_name_or_path setting.

Thanks in advance for your help!

it600-val.mp4

it2000-val.mp4

[GYGWG]

Hi, I meet the exactly same issue. Did you solve it?

[GYGWG]

I set pretrained_model_name_or_path: "./load/zero123/stable_zero123.ckpt" in sugar_static_refine.yaml

[chengu-123]

Hello, I'm a novice in this field. I would like to ask you how you generate a complete multi-view mesh. According to the readme instructions, I only generated a single-view mesh.I don't know how to create multi-view grids.

[xodus777]

@GYGWG Sorry for late reply, but this was all about the mesh initialization stage done before the GS mesh adaptation. I found if initialized mesh is unclear, the sugar refine los snever tends to drop... I haven't found ways to refine this mesh.. probably high-quality image input is the key to solve this issue?

@chengu-123 If you've got these results from the mesh generation stage, maybe this is because of the performance of zero123. If you are facing these issue in sugar refinement stage, I'm also still working for it.. It would be very helpful if you could share your solutions if you've solved the problems