@ichsan2895 no, I did the reconstruction with Colmap and training with Brush. I find Sprirula's reconstruction works very fast as it uses the AMD GPU (unlike Colmap) but is not so precise. So I still prefer Colmap. And for training, these kind of indoor scenes work best with Brush - any other tool I tried has optimizations which cut corners here and there and don't give such a good result.

Aha. You already used spirula-studio for masked people. Do you also train with spirula instead of brush? I just curious with others software output with same dataset such as nerfstudio, spirula, and LFS