This repository is based on PRBonn/OverlapNet. OverlapNet is modified Siamese Network that predicts the overlap and relative yaw angle of a pair of range images generated by 3D LiDAR scans. You may see details from their paper.
This repository use the cosine similarity of two functions that represent two point clouds instead of the overlap value proposed in OverlapNet. Using a similar architecture of OverlapNet to learn and estimate the cosine similarity and relative yaw angle between two point clouds.
- See instructions in the docker folder to build the docker for the environment.
- This repository is tested on Ubuntu 20.04 with Tensorflow-2.4.1
- The cosine similarity of two point clouds are calculated using the code in the cosine_similarity folder which is a simplified code of UMich-CURLY/unified_cvo for generating the cosine similarities over two point clouds.
- To generate groundtruth, training and testing data, run
python3 demo/gen_continuous_groundtruth.py config/demo.yml - To train the network, run
python src/two_heads/training.py config/network.yml - To test the network, run
python src/two_heads/testing.py config/test.yml