Point-Voxel CNN for Efficient 3D Deep Learning

NeurIPS 2019 Zhijian LiuHaotian TangYujun LinSong Han

We present Point-Voxel CNN (PVCNN) for efficient, fast 3D deep learning. Previous work processes 3D data using either voxel-based or point-based NN models... (read more)

PDF Abstract

Results from the Paper


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK RESULT LEADERBOARD
3D Object Detection KITTI Cars Easy val PVCNN AP 84.02 # 1
3D Object Detection KITTI Cars Hard val PVCNN AP 63.81 # 1
3D Object Detection KITTI Cars Moderate val PVCNN AP 71.54 # 1
3D Object Detection KITTI Cyclist Easy val PVCNN AP 81.4 # 1
3D Object Detection KITTI Cyclist Hard val PVCNN AP 56.24 # 1
3D Object Detection KITTI Cyclist Moderate val PVCNN AP 59.97 # 1
3D Object Detection KITTI Pedestrian Easy val PVCNN AP 73.2 # 1
3D Object Detection KITTI Pedestrian Hard val PVCNN AP 56.78 # 1
3D Object Detection KITTI Pedestrian Moderate val PVCNN AP 64.71 # 1
3D Instance Segmentation S3DIS PVCNN++ (1xC) volumetric mIoU 58.98% # 3
mAcc 87.1 # 1
3D Part Segmentation ShapeNet-Part PVCNN volumetric Instance Average IoU 86.2 # 4

Methods used in the Paper


METHOD TYPE
🤖 No Methods Found Help the community by adding them if they're not listed; e.g. Deep Residual Learning for Image Recognition uses ResNet