Skip to content

Latest commit

 

History

History
51 lines (39 loc) · 1.99 KB

README.md

File metadata and controls

51 lines (39 loc) · 1.99 KB
Raw

Robust Object Pose Estimation (ROPE)

This repo stores code used in the paper

Occlusion-Robust Object Pose Estimation with Holistic Representation

Environment

Our system environment is provided in environment.yaml for consideration.

Datasets

The Linemod (lm), Linemod-Occluded (lmo) and YCB-Video (ycbv) datasets can be downloaded from the BOP website. The paths to the datasets should then be specified in the cfg.yaml file.

For better initialisation, the pretrained hrnet backbone file can be downloaded from here.

Usage examples

To train for the lm test set in distrubted mode

python -m torch.distributed.launch --nproc_per_node=<num_gpus_to_use> --use_env main_lm.py --cfg cfg.yaml --obj duck --log_name <name_this_experiment>

To train for the lmo test set in single GPU mode

CUDA_VISIBLE_DEVICES=<which_gpu> python main_lmo.py --cfg cfg.yaml --obj ape --log_name <name_this_experiment>

To load trained model and test on the lmo dataset

CUDA_VISIBLE_DEVICES=<which_gpu> python main_lmo.py --cfg cfg.yaml --obj cat --log_name <which_experiment_to_load> --resume --test-only

To train for the ycbv test set for object 01

python -m torch.distributed.launch --nproc_per_node=<num_gpus_to_use> --use_env main_ycbv.py --cfg cfg.yaml --obj 01 --log_name <name_this_experiment>

To compute AUC for a ycbv test result for object 20

python analysis.py --cfg cfg.yaml --log_name <which_experiment_to_load> --obj 20

Cite this work

@inproceedings{chen2022occlusion,
    Author = {Chen, Bo and Chin, Tat-Jun and Klimavicius, Marius},
    Title = {Occlusion-Robust Object Pose Estimation with Holistic Representation},
    Booktitle = {WACV},
    Year = {2022}
}