How to use the rendering method to improve depth GT

[brucetend]

I just learned that matterport data sets have problems with their depth labels (labeling errors, etc.), and directly using their depth GT to generate BEV semantic segmentation labels will lead to further cumulative errors. Since the ground truth label cannot find a suitable correspondence with your depth map (with a random rotation), could you please teach me, what method should I choose to render the depth GT, If I want to correct it. But I can't see the specific method from the paper. Thank you so much! best regards, Teng