search

shanice-l / gdrnpp_bop2022

PyTorch Implementation of GDRNPP, winner (most of the awards) of the BOP Challenge 2022 at ECCV'22
Apache License 2.0
241 stars 51 forks source link

Where should I get the xyz_crop directory of lmo datasets? #128

Open Aljen-511 opened 5 days ago

Aljen-511 commented 5 days ago

Hi, I'm new in the research of pose estimation, and I downloaded the lmo dataset from huggingface (followed the guidence of official BOP website). The wired thing is, when I tried to train a model on that dataset, I found no xyz_crop directory under lmo's trainset, while the python script ./core/gdrn_modeling/datasets/lm_pbr.py does define a path parameter pointed to that directory (line 354). The training script I used is as below: 

$ ./core/gdrn_modeling/train_gdrn.sh configs/gdrn/lmoPbrSO/convnext_AugCosyAAEGray_DMask_amodalClipBox_lmo/duck.py 0

By the way, It seems that the class LM_PBR_Dataset DO NOT take measures to handle the lmo case, as the PNG format of lmo's rgb images is not considered (line 113). Is it a BUG or did I run the script in a wrong way?

Aljen-511 commented 5 days ago

I also notice that issue #123 has the same problem, but didn't give a solution.

Aljen-511 commented 5 days ago

我在#23我找到类似问题的解决方案。 但是训练gdrn的时候，可以像训练yolox一样注释掉assert assert osp.exists(xyz_path)吗？