strange results when use color point clouds to reconstruct mesh

[NguyenTriTrinh]

Hi ,Zhao, Thanks for sharing your interesting work. I downloaded M3D dataset and tried to run python rgbd2pcd.py to get point clouds using depth maps of the dataset. I set flag label='gt' , but got a strange result.
Here is the front-view rgb.png and the generated rgb_back.png

And here is the mesh I got following your instructions on MeshLab.

Besides, why we use fd = -1.0 * (2.0 * fd - 1.0) rather than fd = 2.0 * fd - 1.0 to transform the fd coordinates? Can you give me some tips to fix that ? really thank.