About test set

[sseunghyuns]

Thanks for your great work.

I have some questions about Wild6D annotated test set.
I'm currently using my model trained on the NOCS dataset for evaluation, and I've encountered some concerns with the RT error results, particularly in the parts marked in red.

Upon visualizing the results with evaluate_wild6d.py, I noticed that several inference outcomes were significantly incorrect.

It turns out that the masks used to generate the point clouds were incorrectly annotated. Including such improperly annotated masks in the evaluation seems to have skewed the results.

Is there something I might have missed while loading the test data? (I'm currently using evaluate_wild6d.py to load the data, and load the depth map and mask with that code.)