Ambiguous Depth GT

Hi, thanks for this great work and dataset. When I use the depth and camera parameters for point cloud projection, I find the depth annotation may be inaccurate in some scenarios. Currently, I found that scene 009850-3_obj_source_left gives an ambiguous point cloud. Here is the visualization. For the rest of the scenarios I tested, the depth seems right.

Reproduce the point-cloud projection for scene 009850-3_obj_source_left using its depth annotation and camera parameters, then compare it with the visualization in the issue. Determine whether the ambiguity comes from the annotation or the projection inputs, and confirm the expected result for this scene before proposing a correction.

Written by the indexing model from the issue text.