Research map / Point clouds / Point cloud classification
GMSF: Global Matching Scene Flow
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| replace coarse-to-fine or recurrent refinement with single-scale one-shot global matching | GMSF · F3D_o | Outliers: 27.4 -> 5.6 | Helped |
| use global-cross transformers rather than no global-cross layers (ablation) | GMSF · F3D_c all | EPE3D: 0.212 -> 0.040 | Helped |
| increase global-cross transformer depth from 2 to 10 layers (ablation) | GMSF · F3D_c all | EPE3D: 0.075 -> 0.040 | Helped |
| increase global-cross transformer depth from 10 to 12 layers (ablation) | GMSF · F3D_c all | EPE3D: 0.040 -> 0.043 | Hurt |
| add a local Point Transformer to DGCNN tokenization (ablation) | GMSF · F3D_c all | EPE3D: 0.052 -> 0.040 | Helped |
| add a local Point Transformer to PointNet tokenization (ablation) | GMSF · F3D_c all | EPE3D: 0.063 -> 0.043 | Helped |
| add a local Point Transformer to MLP tokenization (ablation) | GMSF · F3D_c all | EPE3D: 0.060 -> 0.043 | Helped |
| replace DGCNN with PointNet while retaining local Point Transformer (ablation) | GMSF · F3D_c all | EPE3D: 0.040 -> 0.043 | Mixed |
| reduce feature dimension from 128 to 64 (ablation) | GMSF · F3D_c all | EPE3D: 0.040 -> 0.051 | Hurt |
| reduce feature dimension from 128 to 32 (ablation) | GMSF · F3D_c all | EPE3D: 0.040 -> 0.073 | Hurt |
| use full-dataset Waymo-Open training and evaluation rather than the previously limited 100-sequence setting | GMSF · Waymo-Open | EPE3D | Helped |
Closest papers by meaning
- FlowMamba: Learning Point Cloud Scene Flow with Global Motion Propagation (AAAI 2025 · similarity 0.87)
- Zero-Shot Monocular Scene Flow Estimation in the Wild (CVPR 2025 · similarity 0.86)
- Sparse Global Matching for Video Frame Interpolation with Large Motion (CVPR 2024 · similarity 0.86)
- ZeroFlow: Scalable Scene Flow via Distillation (ICLR 2024 · similarity 0.86)
- DifFlow3D: Toward Robust Uncertainty-Aware Scene Flow Estimation with Iterative Diffusion-Based Refinement (CVPR 2024 · similarity 0.86)
- UFM: A Simple Path towards Unified Dense Correspondence with Flow (NeurIPS 2025 · similarity 0.85)
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.