This repository contains python scripts of paper Match Normalization:Learning-based Point Cloud Registration for 6D Object Pose Estimation in the Real World.
-
BOP Toolkit before running the code please setup the path for loading the BOP Benchmark data.
-
conda env create -f environment.yml
python scratch.py --mode='ori' --bop_dataset='tudl'
[mAP Metric]
python scratch.py --mode='ckpt' --bop_dataset='tudl' --exp_name='bpnet_tudl'
python scratch.py --mode='ckpt' --bop_dataset='lm' --exp_name='bpnet_lm'
python scratch.py --mode='ckpt' --bop_dataset='lmo' --exp_name='bpnet_lm'
[BOP Metric]
python eval_bop19.py --result_filenames='zheng-bpnet_tudl-test.csv'
python eval_bop19.py --result_filenames='zheng-bpnet_lm-test.csv'
python eval_bop19.py --result_filenames='zheng-bpnet_lmo-test.csv'
-
pretrained models are saved at
./ckpt
-
bop benchmark result files are saved at
./bop_res
@inproceedings{Zheng2022,
title={Learning-based Point Cloud Registration for 6D Object Pose Estimation in the Real World},
author={Dang, Zheng and Wang, Lizhou and Guo, Yu and Salzmann, Mathieu},
booktitle = {European Conference on Computer Vision(ECCV) 2022},
month = {October},
year={2022}
}
MIT License