Research map / Point clouds / Point cloud classification
PointTree: Transformation-Robust Point Cloud Encoder with Relaxed K-D Trees
Jun-Kun Chen, Yu-Xiong Wang
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 |
|---|---|---|---|
| use relaxed PCA-based K-D tree instead of the original axis-aligned K-D tree (ablation) | PointTree Def · ModelNet40 affine (with PA) | overall accuracy (%): 82.7 -> 68.8 when replaced with original K-D tree | Improved |
| merge child info with pointwise maximum instead of concatenate-MLP (ablation) | PointTree RNS · ModelNet40 affine (with PA) | overall accuracy (%): 84.1 -> 79.3 with concatenate-MLP | Improved |
| add the learnable alignment network (ablation) | PointTree Def · ModelNet40 affine (with PA) | overall accuracy (%): 82.4 -> 82.7 | Improved |
| use stronger K-D tree alignment network (KA) rather than T-Net (Def) (ablation) | PointTree · ModelNet40 affine (with PA) | overall accuracy (%): 82.7 -> 83.4 | Improved |
| use ResNet-style stacked encoder (RNS) rather than Def (ablation) | PointTree · ModelNet40 affine (with PA) | overall accuracy (%): 82.7 -> 84.1 | Improved |
| add PCA pre-alignment (ablation) | PointTree RNS · ModelNet40 affine | overall accuracy (%): 73.1 -> 84.1 | Improved |
| add PCA pre-alignment (ablation) | PointTree Def · ModelNet40 affine | overall accuracy (%): 70.2 -> 82.7 | Improved |
| add PCA pre-alignment to baselines | PointNet, DGCNN, PointMLP and other baselines · ModelNet40 affine | overall accuracy (%): at least 10% improvement in accuracy | Improved |
| add pre-alignment to baselines | DGCNN · ModelNet40 projective | overall accuracy (%): 6.2 -> 47.3 | Improved |
| replace PointNet-style MLP/ball clustering with relaxed K-D tree design (PointTree RNS vs 3DContextNet) | PointTree RNS · ModelNet40 affine | overall accuracy (%): 76.7 -> 84.1 with PA; 37.1 -> 73.1 without PA | Improved |
| PD-Net coordinate-dependent aggregation vs PointTree aggregation | PD-Net · ModelNet40 affine (with PA) | overall accuracy (%): 62.0 vs 84.1 | Hurt |
| apply PointTree encoder to ShapeNetPart part segmentation | PointTree RNS · ShapeNetPart affine (with PA) | class mIoU (%): 71.7 -> 78.7 vs GDANet | Improved |
Closest papers by meaning
- Multiresolution Tree Networks for 3D Point Cloud Processing (ECCV 2018 · similarity 0.88)
- AdaPoinTr: Diverse Point Cloud Completion With Adaptive Geometry-Aware Transformers (TPAMI 2023 · similarity 0.88)
- PoinTr: Diverse Point Cloud Completion With Geometry-Aware Transformers (ICCV 2021 · similarity 0.88)
- Frozen CLIP Transformer Is an Efficient Point Cloud Encoder (AAAI 2024 · similarity 0.88)
- Unsupervised Point Cloud Pre-Training via Occlusion Completion (ICCV 2021 · similarity 0.87)
- TopNet: Structural Point Cloud Decoder (CVPR 2019 · 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.
