Research map / Point clouds / Point cloud classification
Point MixSwap: Attentional Point Cloud Mixing via Swapping Matched Structural Divisions
Ardian Umam, Cheng-Kun Yang, Yung-Yu Chuang, Jen-Hui Chuang, Yen-Yu Lin
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add feature-level Point MixSwap instead of input-level (ablation) | DGCNN · M40 (20%) | accuracy: 89.5 -> 89.7 | Improved |
| combine input-level and feature-level mixup instead of feature-level alone (ablation) | DGCNN · M40 (20%) | accuracy: 89.7 -> 89.5 | No clear effect |
| add Point MixSwap on top of conventional data augmentation (ablation) | DGCNN · M40 (20%) | accuracy: 88.7 -> 91.3 | Improved |
| add attention encoder-decoder blocks without mixup to CDA (ablation) | DGCNN · M40 (20%) | accuracy: 88.7 -> 88.8 | No clear effect |
| replace learned divisions with trivial horizontal/vertical/random division (ablation) | DGCNN + CDA + Point MixSwap · M40 (20%) | accuracy: 91.3 -> 89.2, 89.0, 88.9 | Hurt |
| add Point MixSwap to conventional augmentation (ablation) | DGCNN · M10 (20%) | accuracy: 93.2 -> 94.6 | Improved |
| add Point MixSwap to conventional augmentation (ablation) | DGCNN · SON (20%) | accuracy: 73.0 -> 76.3 | Improved |
| increase number of divisions in input-level mixup from 3 to 5 (ablation) | DGCNN · SON (20%) | accuracy: 76.1 -> 75.5 | No clear effect |
| add principal axis alignment (PAA) to Point MixSwap (ablation) | DGCNN · RM40 (20%) | accuracy: 89.3 -> 90.1 | Improved |
| add Point MixSwap to full training data | DGCNN · M40 (100%) | accuracy: 92.7 -> 93.5 | Improved |
| add Point MixSwap with 50% data vs baseline on full data | PointNet · M40 | accuracy: 88.6 -> 88.7 | Improved |
| use Point MixSwap instead of RSMix | DGCNN · M10 (100%) | accuracy: 95.9 -> 96.0 | No clear effect |
Closest papers by meaning
- PointPatchMix: Point Cloud Mixing with Patch Scoring (AAAI 2024 · similarity 0.92)
- PointMixup: Augmentation for Point Clouds (ECCV 2020 · similarity 0.90)
- SageMix: Saliency-Guided Mixup for Point Clouds (NeurIPS 2022 · similarity 0.89)
- On Geometry-Enhanced Parameter-Efficient Fine-Tuning for 3D Scene Segmentation (NeurIPS 2025 · similarity 0.88)
- Rethinking Point Cloud Data Augmentation: Topologically Consistent Deformation (ICML 2025 · similarity 0.87)
- LinNet: Linear Network for Efficient Point Cloud Representation Learning (NeurIPS 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
