RelPoseNet

A PyTorch version of the ego-motion estimation pipeline proposed in our work. The official implementation (in Lua) is available at https://github.com/AaltoVision/camera-relocalisation

Evaluation on the 7-Scenes dataset

scene	Lua	PyTorch (this repo)
Chess	0.13m, 6.46deg	0.12m, 7.10deg
Fire	0.26m, 12.72deg	0.26m, 12.45deg
Heads	0.14m, 12.34deg	0.14m, 11.72deg
Office	0.21m, 7.35deg	0.20m, 9.23deg
Pumpkin	0.24m, 6.35deg	0.21m, 8.10deg
Red Kitchen	0.24m, 8.03deg	0.23m, 8.82deg
Stairs	0.27m, 11.82deg	0.27m, 11.66deg
Average	0.21m, 9.30deg	0.20m, 9.87deg

Installation

• create and activate conda environment with Python 3.x

conda create -n my_fancy_env python=3.7
source activate my_fancy_env

• install all dependencies by running the following command:

pip install -r requirements.txt

Evaluation and Training

Evaluation and training have been performed on the 7-Scenes dataset available here. Important!!! The images have to be resized such that the smaller dimension is 256 and the aspect ratio is intact. This could be done using the following command: find . -name "*.color.png" | xargs -I {} convert {} -resize "256^>" {}

Evaluation

• download an archive with the model snapshot and unpack it to the working directory
• navigate to RelPoseNet/experiments and modify the main config file configs/main.yaml. Here, you need to change work_dir and datasets_home_dir
• modify img_path in the configs/experiment/7scenes.yaml config file. Where img_path is the path with resized images of the 7-Scenes dataset
• run main.py from experiments path
• once evaluated, the script creates a text file with relative camera poses located at ${experiment.experiment_params.output.home_dir}/est_rel_poses.txt
• in order to predict absolute poses, run MATLAB and open experiments/seven_scenes/filter_pose.m
• modify line 17 by providing the text file with estimated relative poses
• if everything goes fine, one should get localization performance presented in the table above.

Training

• modify a config file RelPoseNet/configs/main.yaml by changing work_dir, img_dir, and out_dir
• to perform training, run RelPoseNet/main.py

License

Our code is released under the Creative Commons BY-NC-SA 3.0, available only for non-commercial use.

How to cite

If you use this project in your research, please cite:

@inproceedings{Laskar2017PoseNet,
      title = {Camera relocalization by computing pairwise relative poses using convolutional neural network},
      author = {Laskar, Zakaria and Melekhov, Iaroslav and Kalia, Surya and Kannala, Juho},
       year = {2017},
       booktitle = {Proceedings of the IEEE International Conference on Computer Vision Workshops}
}

@inproceedings{Melekhov2017RelPoseNet,
      title = {Camera relocalization by computing pairwise relative poses using convolutional neural network},
      author = {Melekhov, Iaroslav and Ylioinas, Juha and Kannala, Juho and Rahtu, Esa},
       year = {2017},
       booktitle = {International Conference on Advanced Concepts for Intelligent Vision Systems}
}