NicolaBernini
commented
4 years ago Comparison with other Approaches
- Other approaches working on Point Cloud typically require a pre-processing step to transform it into 3D Voxel or RGB-like Image hence a
WxHxCTensor but this kind of strategy has some limitations- it increases the Data to Process as the target data structures are more dense than the original Point Cloud
- it requires some inductive bias at the pre-processing step (manually engineered features, manually set hyperparam) which can be suboptimal or at least not as good as the data learned ones
Overview
PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation
Arxiv: https://arxiv.org/abs/1612.00593
GitHub: https://github.com/charlesq34/pointnet
Analysis of PointNet : a DNN designed to work directly on Point Cloud Data to solve tasks like Classification, Parts Segmentation and Semantic Segmentation