Is it possible to accelerate the training by using an existing rough mesh?

[Vilour]

Say if we have an existing rough mesh from Colmap, can we fit the SDF model first, and then train the whole pipeline? Can we accelerate the training procedure by using prior knowledge like this?

[niujinshuchong]

Hi, I think this is definitely possible to speed up the training with colmap reconstruction. but unfortunate our current repo don't support this. Alternatively, you could try to render the depth map from the colmap's mesh and then use the depth map as in the RGBD data example https://github.com/autonomousvision/sdfstudio/blob/master/docs/sdfstudio-examples.md#neus-rgbd-on-the-synthetic-neural-rgbd-dataset