Dense Pose Object Detector (DPOD)

PyTorch implementation of the DPOD detector based on ICCV 2019 paper "DPOD: 6D Pose Object Detector and Refiner", cf. References below. [Full paper]

Dependencies

• PyTorch (torch) (BSD License: https://github.com/pytorch/pytorch/blob/master/LICENSE)
• OpenCV (cv2) (BSD License: https://opencv.org/license/)
• NumPy (BSD License: https://numpy.org/doc/stable/license.html)
• SciPy (BSD License: https://www.scipy.org/scipylib/license.html)
• scikit-learn (BSD License: https://github.com/scikit-learn/scikit-learn/blob/master/COPYING)
• pandas (BSD License: https://github.com/pandas-dev/pandas/blob/master/LICENSE)
• plyfile (GNU General Public License v3 or later (GPLv3+): https://github.com/dranjan/python-plyfile/blob/master/COPYING)
• PyYAML (MIT License: https://github.com/yaml/pyyaml/blob/master/LICENSE)

Setting up the environment

Set up a virtual environment using:

conda env create -n dpod -f environment.yml
conda activate dpod

Usage

To test the code activate the created virtual environment and execute the following command:

python main.py config.ini -t

For training the model run:

python main.py config.ini

Datasets

Mini versions of the training and test datasets as well as the 3D models from the LineMOD dataset are located in the db_mini folder.

• models - 3D models from the LineMOD dataset
• models_uv - 3D models with UV texture
• test - RGB test images from the LineMOD dataset
• train - rendered train patch images, i.e. rgb, correspondences (uv or uvw), normals, and sample backgrounds from MS COCO

Pretrained networks for LineMOD dataset trained on synthetic renderings can be found under the following link.

References

DPOD: 6D Pose Object Detector and Refiner (ICCV 2019)

Sergey Zakharov*, Ivan Shugurov*, Slobodan Ilic

@inproceedings{dpod,
author = {Sergey Zakharov and Ivan Shugurov and Slobodan Ilic},
title = {DPOD: 6D Pose Object Detector and Refiner},
booktitle = {International Conference on Computer Vision (ICCV)},
month = {October},
year = {2019}
}