custom dataset with depth2map

[wangrui5781]

I replaced colmap with hloc as your second "to do" in readme, and got a little optimization.
But when I use casmvsnet and depth2dmap to generate *.dmap instead of openmvs, I got a chaotic result. The target is a person but a cube is generated. I am wondering how your model is trained and how can I get correct dmap.