Wuuu3511 / GoMVS

[CVPR'24]🦿GoMVS: Geometrically Consistent Cost Aggregation for Multi-View Stereo

Home Page:https://wuuu3511.github.io/gomvs/

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

The normal maps for TNT

GuangyuWang99 opened this issue · comments

Hi,

Thanks for the excellent work! However, I found all provided monocular normal data (e.g., in this link) seems to be broken. Could you help check it and upload it again?

Thanks in advance!

Besides, I'm wondering whether this method is robust to the normal priors. That is, once the model is pretrained using Ominidata normals, will the results deteriorate when inference with normals from other high quality monocular models (e.g., DepthAnything, Marigold, DSINE), due to sorts of distribution shift?

Besides, I'm wondering whether this method is robust to the normal priors. That is, once the model is pretrained using Ominidata normals, will the results deteriorate when inference with normals from other high quality monocular models (e.g., DepthAnything, Marigold, DSINE), due to sorts of distribution shift?

Thank you for your attention to our work.

I have updated the Baidu Cloud link for the TNT normal data: https://pan.baidu.com/s/1Vep8lANnx3UqBNy4ZMXatQ?pwd=et97.

We have not tested normals generated by other monocular networks. Perhaps similar normal prediction methods could yield similar results.