Research map / Point clouds / Point cloud classification
Multiresolution Tree Networks for 3D Point Cloud Processing
Matheus Gadelha, Rui Wang, Subhransu Maji
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 multiresolution convolution to the encoder (vs single resolution) (ablation) | MRTNet encoder · ModelNet40 classification | accuracy (%): 89.3 -> 91.7 | Improved |
| add data augmentation to the single-resolution model, kd-tree (ablation) | single-resolution encoder · ModelNet40 classification | accuracy (%): 86.2 -> 89.3 | Improved |
| use rp-tree instead of kd-tree ordering without augmentation (ablation) | single-resolution encoder · ModelNet40 classification | accuracy (%): 86.2 -> 87.4 | Improved |
| reduce filters by 4 to match parameter count (ablation) | MRTNet encoder · ModelNet40 classification | accuracy (%): 91.7 -> 91.7 | No clear effect |
| increase points from 1K to 4K | MRTNet · ModelNet40 classification | accuracy (%): 91.2 -> 91.7 | Improved |
| condition convolutions on splitting axes (KDNet) vs single-resolution baseline (ablation) | KDNet-like single resolution · ModelNet40 classification | accuracy (%): 1.3% less accuracy without it | Improved |
| add multiresolution decoder instead of fully connected decoder (ablation) | MRTNet decoder · ShapeNet shape inference | pred->GT error (x100): 1.824 -> 1.559 | Improved |
| add multiresolution decoder instead of fully connected decoder (ablation) | MRTNet decoder · ShapeNet shape inference | GT->pred error (x100): 2.297 -> 1.529 | Improved |
| add multiresolution convolution to the decoder instead of single-resolution (ablation) | MRTNet decoder · ShapeNet shape inference | pred->GT error (x100): 1.708 -> 1.559 | Improved |
| add multiresolution convolution to the decoder instead of single-resolution (ablation) | MRTNet decoder · ShapeNet shape inference | GT->pred error (x100): 1.831 -> 1.529 | Improved |
| use MRTNet instead of voxel-based 3D-R2N2 (5 views) | MRTNet · ShapeNet shape inference | pred->GT error (x100): 2.588 -> 1.559 | Improved |
| use MRTNet instead of Lin et al. multi-view method | MRTNet · ShapeNet shape inference | GT->pred error (x100): 1.701 -> 1.529 | Mixed |
Closest papers by meaning
- Unsupervised 3D Learning for Shape Analysis via Multiresolution Instance Discrimination (AAAI 2021 · similarity 0.90)
- TopNet: Structural Point Cloud Decoder (CVPR 2019 · similarity 0.88)
- PointTree: Transformation-Robust Point Cloud Encoder with Relaxed K-D Trees (ECCV 2022 · similarity 0.88)
- EditVAE: Unsupervised Parts-Aware Controllable 3D Point Cloud Shape Generation (AAAI 2022 · similarity 0.88)
- SoftPool++: An Encoder–Decoder Network for Point Cloud Completion (IJCV 2022 · similarity 0.87)
- Learning to Tessellate: Point Cloud Generation via Recursive Spectral Partitioning (ECCV 2026 · 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.
