Research map / Point clouds / Point cloud classification
Rethinking Serialization in Linear 3D Vision: Decoupling Anisotropic Geometry from Isotropic Semantics
YinYun Yan, Liping zhang, Tingran Wang, Jiaxin Deng, Changshuo Wang, Limin Jiang, Shanwei Gao, Xin Ning
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 ellipsoidal LAGM encoding over spherical (ablation) | Sphere + MLP · S3DIS Area 5 | mIoU (%): 73.48 -> 74.44 | Improved |
| add GISA global aggregation over MLP (ablation) | Sphere + MLP · S3DIS Area 5 | mIoU (%): 73.48 -> 81.94 | Improved |
| add LAGM on top of GISA (ablation) | Sphere + GISA · S3DIS Area 5 | mIoU (%): 81.94 -> 82.62 | Improved |
| use Identity mode instead of Hilbert serialization (ablation) | AnIsoNet · S3DIS Area 5 | mIoU (%): 74.46 -> 82.62 | Improved |
| use Identity mode instead of Morton serialization (ablation) | AnIsoNet · S3DIS Area 5 | mIoU (%): 74.68 -> 82.62 | Improved |
| use Morton mode instead of Identity (ablation) | AnIsoNet · ScanObjectNN (PB T50 RS) | OA (%): 92.51 -> 94.21 | Improved |
| use Morton mode instead of Hilbert (ablation) | AnIsoNet · ScanObjectNN (PB T50 RS) | OA (%): 93.86 -> 94.21 | Improved |
| perturb inference input order (reverse, Morton, Hilbert, random, chunk shuffle) (ablation) | AnIsoNet Identity mode · ScanNetV2 val | mIoU (%): 78.47 -> 78.43 | No clear effect |
| replace PCM multi-view serialization with single-pass AnIsoNet | PCM · S3DIS Area 5 | mIoU (%): 79.6 -> 82.62 | Improved |
| replace PTv3 with AnIsoNet | PTv3 · S3DIS Area 5 | mIoU (%): 73.4 -> 82.62 | Improved |
| replace PTv3 with AnIsoNet without pre-training | PTv3 · ScanNetV2 val | mIoU (%): 77.5 -> 78.52 | Improved |
| AnIsoNet versus Sonata (dec.) with external pre-training | Sonata (dec.) · ScanNetV2 val | mIoU (%): 79.1 -> 78.52 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 4 of 4 on most shared results.
- PointMamba — PointMamba: A Simple State Space Model for Point Cloud Analysis (NeurIPS 2024) · better on 3 of 3 shared results
- PointRWKV — PointRWKV: Efficient RWKV-Like Model for Hierarchical Point Cloud Learning (AAAI 2025) · better on 1 of 1 shared result
- StruMamba3D — StruMamba3D: Exploring Structural Mamba for Self-supervised Point Cloud Representation Learning (ICCV 2025) · better on 1 of 1 shared result
- OctFormer — OctFormer: Efficient Octree-Based Transformer for Point Cloud Compression with Local Enhancement (AAAI 2023) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Large Spatial Model: End-to-end Unposed Images to Semantic 3D (NeurIPS 2024 · similarity 0.85)
- LinNet: Linear Network for Efficient Point Cloud Representation Learning (NeurIPS 2024 · similarity 0.85)
- On Geometry-Enhanced Parameter-Efficient Fine-Tuning for 3D Scene Segmentation (NeurIPS 2025 · similarity 0.85)
- G^3-LQ: Marrying Hyperbolic Alignment with Explicit Semantic-Geometric Modeling for 3D Visual Grounding (CVPR 2024 · similarity 0.85)
- Neural Isometries: Taming Transformations for Equivariant ML (NeurIPS 2024 · similarity 0.85)
- FrameVGGT: Coherence-Preserving Memory for Bounded Streaming Geometry (NeurIPS 2026 · similarity 0.84)
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.
