Running evaluation on BOP datasets

[OrestisVaggelis]

I’m attempting to run the pose estimation evaluation on all standard BOP datasets using the current datareader.py implementation with minimal modifications to run_linemod.py. While I achieve very reasonable (in fact higher than reported) results on LMO and YCBV, the results for ICBIN, TLESS, and TUDL are unexpectedly low.

Are there any additional changes needed in run_linemod.py, datareader.py, or elsewhere to properly handle ICBIN, TLESS, and TUDL?

[wenbowen123]

Hi, as mentioned in the readme, the released weights are not exactly the same as what we used to run the experiments (excluding the diffusion augmented data, due to licensing issues). It might worth checking how do those failure cases look like on "ICBIN, TLESS, and TUDL" to understand why.