Homography-Based Loss Function for Camera Pose Regression

In this repository, we share our implementation of several camera pose regression loss functions in a simple end-to-end network similar to PoseNet. We implemented the homography-based loss functions introduced in our paper alongside PoseNet, Homoscedastic, Geometric and DSAC loss functions. We provide the code to train and test the network on Cambridge, 7-Scenes and custom COLMAP datasets.

Our paper Homography-Based Loss Function for Camera Pose Regression is published in IEEE Robotics and Automation Letters 2022.

Convergence of our proposed Homography loss

We show other losses convergence on our YouTube channel.

Installation

COLMAP dependancy

This code relies on COLMAP for loading COLMAP models. To satisfy this dependancy, simply run:

git submodule update --init

Python environment setup

We share an Anaconda environment that can be easily installed by running:

conda env create -f environment.yml

Anaconda is easy to install and benefits from a lighter implementation named Miniconda.
Once the environment is installed you can activate it by running:

conda activate homographyloss

Dataset setup

Have a look at the datasets folder to setup the datasets.

Run relocalization

The script main.py trains the network on a given scene and logs the performance of the model on the train set. It requires one positional argument: the path to the scene on which to train the model. For example, for training the model on the ShopFacade scene, simply run:

python main.py datasets/ShopFacade

Let's say you have a custom dataset in datasets/mydataset with the structure defined in datasets:

• mydataset
  • images
    • frame001.jpg
    • frame002.jpg
    • frame003.jpg
    • ...
  • cameras.bin
  • images.bin
  • points3D.bin
  • list_db.txt
  • list_query.txt

Then you might run the script on your custom dataset:

python main.py datasets/mydataset

Other available training options can be listed by running python main.py -h.

Monitor training and test results

Training and test metrics are saved in a logs directory. One can monitor them using tensorboard. Simply run in a new terminal:

tensorboard --logdir logs

All estimated poses are also saved in a CSV file in logs/[scene]/[loss]/epochs_poses_log.csv. For each epoch, each image and each set, we save the estimated pose in the following format:

• w_t_chat is the camera-to-world translation of the image.
• chat_q_w is the world-to-camera quaternion representing the rotation of the image.

Acknowledgements

This work was supported by Ifremer, DYNI team of LIS laboratory and COSMER laboratory.

License

This code is released under the LGPLv3 licence. Please have a look at the licence file at the repository root.

Citation

If you use this work for your research, please cite:

@article{boittiaux2022homographyloss,
  author={Boittiaux, Cl\'ementin and
          Marxer, Ricard and
          Dune, Claire and
          Arnaubec, Aur\'elien and
          Hugel, Vincent},
  journal={IEEE Robotics and Automation Letters},
  title={Homography-Based Loss Function for Camera Pose Regression},
  year={2022},
  volume={7},
  number={3},
  pages={6242-6249},
}