PointCNN: Convolution On $\mathcal{X}$-Transformed Points

We present a simple and general framework for feature learning from point clouds. The key to the success of CNNs is the convolution operator that is capable of leveraging spatially-local correlation in data represented densely in grids (e.g. images). However, point clouds are irregular and unordered, thus directly convolving kernels against features associated with the points, will result in desertion of shape information and variance to point ordering. To address these problems, we propose to learn an $\mathcal{X}$-transformation from the input points, to simultaneously promote two causes. The first is the weighting of the input features associated with the points, and the second is the permutation of the points into a latent and potentially canonical order. Element-wise product and sum operations of the typical convolution operator are subsequently applied on the $\mathcal{X}$-transformed features. The proposed method is a generalization of typical CNNs to feature learning from point clouds, thus we call it PointCNN. Experiments show that PointCNN achieves on par or better performance than state-of-the-art methods on multiple challenging benchmark datasets and tasks.

PDF Abstract

Results from the Paper

Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
3D Part Segmentation IntrA PointCNN IoU (V) 93.59 # 4
DSC (V) 96.62 # 4
IoU (A) 74.11 # 5
DSC (A) 81.74 # 5
Few-Shot 3D Point Cloud Classification ModelNet40 5-way (10-shot) PointCNN Overall Accuracy 65.41 # 18
Standard Deviation 8.9 # 17
3D Instance Segmentation S3DIS PointCNN mIoU 65.39% # 1
mAcc 75.61 # 1
3D Point Cloud Classification ScanObjectNN PointCNN Overall Accuracy 78.5 # 52
Mean Accuracy 75.1 # 25
OBJ-BG (OA) 86.1 # 14
OBJ-ONLY (OA) 85.5 # 14
3D Part Segmentation ShapeNet-Part PointCNN Class Average IoU 84.6 # 11
Instance Average IoU 86.14 # 32