PyTorch implementation of D3Feat for CVPR'2020 Oral paper "D3Feat: Joint Learning of Dense Detection and Description of 3D Local Features", by Xuyang Bai, Zixin Luo, Lei Zhou, Hongbo Fu, Long Quan and Chiew-Lan Tai. D3Feat is also available in Tensorflow.
This paper focus on dense feature detection and description for 3D point clouds in a joint manner. If you find this project useful, please cite:
@article{bai2020d3feat,
title={D3Feat: Joint Learning of Dense Detection and Description of 3D Local Features},
author={Xuyang Bai, Zixin Luo, Lei Zhou, Hongbo Fu, Long Quan and Chiew-Lan Tai},
journal={arXiv:2003.03164 [cs.CV]},
year={2020}
}
A successful point cloud registration often lies on robust establishment of sparse matches through discriminative 3D local features. Despite the fast evolution of learning-based 3D feature descriptors, little attention has been drawn to the learning of 3D feature detectors, even less for a joint learning of the two tasks. In this paper, we leverage a 3D fully convolutional network for 3D point clouds, and propose a novel and practical learning mechanism that densely predicts both a detection score and a description feature for each 3D point. In particular, we propose a keypoint selection strategy that overcomes the inherent density variations of 3D point clouds, and further propose a self-supervised detector loss guided by the on-the-fly feature matching results during training. Finally, our method achieves state-of-the-art results in both indoor and outdoor scenarios, evaluated on 3DMatch and KITTI datasets, and shows its strong generalization ability on the ETH dataset. Towards practical use, we show that by adopting a reliable feature detector, sampling a smaller number of features is sufficient to achieve accurate and fast point cloud alignment.
-
Create the environment and install the required libaries:
conda env create -f environment.yml -
Compile the C++ extension module for python located in
cpp_wrappers. Open a terminal in this folder, and run:sh compile_wrappers.sh
We only support 3DMatch dataset currently. Please look for the detailed instructions to download 3DMatch dataset in Original D3Feat Repo. To train the network, please run
python train.pyThe configuration can be changed in config.py. The snapshot and tensorboard file be saved in snapshot/ and tensorboard/. The testing can be done by runnning
python test.py --chosen_snapshot [timestr of the model]Then the geometric registration result will be saved in geometric_registration/.
We would like to thank the open-source code of KPConv.