PointCloud-C is the very first test-suite for point cloud robustness analysis under corruptions.
- Two sets: ModelNet-C for point cloud classification and ShapeNet-C for part segmentation.
- Real-world corruption sources, ranging from object-, senor-, and processing-levels.
- Seven types of corruptions, each with five severity levels.
- Benchmark with more than 20 point cloud recognition algorithms.
- Methods ranging from architecture design, augmentations, and pre-training.