RSKDD-Net: Random Sample-based Keypoint Detector and Descriptor

Authors: Fan Lu, Guang Chen, Yinlong Liu, Zhongnan Qu, Alois Knoll

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on two large scale outdoor Li DAR datasets show that the proposed RSKDD-Net achieves state-of-the-art performance with more than 15 times faster than existing methods.
Researcher Affiliation Academia Fan Lu Tongji University lufan@tongji.edu.cn Guang Chen Tongji University guangchen@tongji.edu.cn Yinlong Liu Technische Universität München Yinlong.Liu@tum.de Zhongnan Qu ETH Zurich quz@ethz.ch Alois Knoll Technische Universität München knoll@in.tum.de
Pseudocode No The paper describes the methods in text and uses diagrams (Fig. 1) to illustrate network architecture, but it does not contain structured pseudocode or algorithm blocks.
Open Source Code Yes Our code is available at https://github.com/ispc-lab/RSKDD-Net.
Open Datasets Yes We evaluate our proposed RSKDD-Net on two large scale outdoor Li DAR datasets, namely KITTI Odometry Dataset [37] (KITTI dataset) and Ford Campus Vision and Lidar Dataset [38] (Ford dataset).
Dataset Splits Yes KITTI dataset provides 11 sequences (00-10) with ground truth vehicle poses and we use Sequence 00 to train, Sequence 01 for validation and the others for testing
Hardware Specification Yes The network is trained on NVIDIA GeForce 1080Ti and evaluated on a PC with Intel i7-9750H and NVIDIA GeForce RTX 2060.
Software Dependencies No The paper mentions "implemented using PyTorch [40]", but does not specify a version number for PyTorch or any other software dependencies.
Experiment Setup Yes The dilation ratio αd is set to 2 and the number of neighbor points is set to 128. The network is implemented using PyTorch [40]. We use SGD as the optimizer with learning rate of 0.001 and momentum of 0.9. Temperature t in matching loss is set to 0.1.