search

NVlabs / neuralangelo

Official implementation of "Neuralangelo: High-Fidelity Neural Surface Reconstruction" (CVPR 2023)
https://research.nvidia.com/labs/dir/neuralangelo/
Other
4.31k stars 387 forks source link

Bad Result on My Data #124

Open Adolfhill opened 11 months ago

Adolfhill commented 11 months ago

Hello, thanks for your work! I am trying to run neuralanglo on my data, but I get an extremely weird result. I checked the state of training and found that it seems all right. Here are some details. Could you please help me correct my results?

the result I got:

微信图片_20230925133150

the state of training:

微信图片_20230925133325 微信图片_20230925133343 微信图片_20230925133410 微信图片_20230925133258

the result of visualize_colmap.ipynb:

微信图片_20230925133518

mli0603 commented 11 months ago

Hi @Adolfhill

If everything is set up correctly and you have increased the batch size, it is likely that this capture is quite hard to reconstruct for the current implementation. It looks like the trajectory is purely forward without much overlapping between the views.

Adolfhill commented 11 months ago

So I should walk around there a few more times, right? According to your experience, how many circles will produce better results in this environment? thank you for your help!

mli0603 commented 11 months ago

Hi @Adolfhill

Neuralangelo and most NeRF methods will require dense camera coverage similar to taking photos of an object in a hemisphere. In your case, walking circularly around the scene will generally make the results better.

Adolfhill commented 11 months ago

Thanks for your help! I will try again based on your suggestion. Do you have any example videos of outdoor scenes that I can refer to?

raficabral commented 3 months ago

@Adolfhill did you have better results?