sailor-z / LocPoseNet

Office implementation of "LocPoseNet: Robust Location Prior for Unseen Object Pose Estimation", 3DV 2024

LocPoseNet: Robust Location Prior for Unseen Object Pose Estimation

This repository is the official implementation of the LocPoseNet: Robust Location Prior for Unseen Object Pose Estimation, 3DV 2024.

Setup Dependencies

conda create -n LocPoseNet
conda activate LocPoseNet
bash ./install.sh

Data Preparation

Please refer to the instructions provided in Gen6D for downloading and preprocessing the data. You may need to adjust the values for DATA_PATH in dataset/database.py, dataset/train_dataset.py, and dataset/train_meta_info.py, to match the actual directory path of your data.

Test pretrained model

We provide a pretrained detector. Run the following evaluation to get the results on LINEMOD and GenMOP:

python ./eval_detector.py

The results should be comparable with those reported in our paper.

Train from scratch

Run the following script to train the model:

python ./train_model.py

Citation

If you find the project useful, please consider citing:

@article{zhao2024locposenet,
  title={Locposenet: Robust location prior for unseen object pose estimation},
  author={Zhao, Chen and Hu, Yinlin and Salzmann, Mathieu},
  journal={International Conference on 3D Vision},
  year={2024}
}