autonomous driving

1. SECOND & History paper:SECOND: Sparsely Embedded Convolutional Detection graph TB VoxelNet --> SECOND --> PointPillar Related paper: VoxelNet: stacked VFE(voxel feature encoding module) + 3D Conv RPN This paper: 3D Conv --> Sparse Conv --> improved sparse conv angle loss regression approach We apply sparse convolution in LiDAR-based object detection, thereby greatly increasing the speeds of training and inference. We propose an improved method of sparse convolution that allows it to run faster....

1. VoxelNet & History paper:VoxelNet: End-to-End Learning for Point Cloud Based 3D Object Detection Related works: Several methods project point clouds into a perspective view and apply image-based feature extraction techniques.Other approaches rasterize point clouds into a 3D voxel grid and encode each voxel with handcrafted feature PointNet:an end-to-end deep neural network that learns point-wise features directly from point clouds,moving from hand-crafted features to learned features. RPN: Region Proposal Network This Paper: Structure: VFE layer: Voxel Feature Encoding layer enables inter-point interaction within a voxel; Pipeline: Divides point clouds into equally spaced 3D voxels and encodes each voxel with stacked VFE layers; 3D Conv RPN Advantages:This efficient algorithm benefits both from the sparse point structure and efficient parallel processing on the voxel grid....