Research map / Learning theory and algorithms / Optimal transport
FLOT: Scene Flow on Point Clouds guided by Optimal Transport
Gilles Puy, Alexandre Boulch, Renaud Marlet
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 residual flow refinement network h (ablation) | FLOT (K=1) · FlyingThings3D_s validation | EPE: 0.1721 -> 0.0785 | Improved |
| add residual flow refinement network h (ablation) | FLOT (K=1) · FlyingThings3D_o validation | EPE: 0.3412 -> 0.1798 | Improved |
| add residual flow refinement network h (ablation) | FLOT (K=1) · FlyingThings3D_p validation (perfect world) | EPE: 0.0010 -> 0.0011 | No clear effect |
| add optimal transport mass regularisation (K=1) over FLOT0 (ablation) | FLOT · FlyingThings3D_s validation (with refinement) | EPE: 0.0811 -> 0.0785 | Improved |
| add optimal transport mass regularisation (K=1) over FLOT0 (ablation) | FLOT · FlyingThings3D_p validation (perfect world) | EPE: 0.0026 -> 0.0011 | Improved |
| increase Sinkhorn iterations K from 1 to 3 (ablation) | FLOT · FlyingThings3D_p validation (perfect world) | EPE: 0.0011 -> 0.0009 | Improved |
| increase Sinkhorn iterations K from 3 to 5 (ablation) | FLOT · FlyingThings3D_s validation (with refinement) | EPE: 0.0786 -> 0.0798 | Hurt |
| increase Sinkhorn iterations K from 3 to 5 (ablation) | FLOT · FlyingThings3D_s validation (with refinement) | AS (%): 51.06 -> 49.77 | Hurt |
| increase Sinkhorn iterations K from 1 to 3 (ablation) | FLOT · FlyingThings3D_o test | EPE: 0.156 -> 0.161 | Hurt |
| add optimal transport mass regularisation (K=1) over FLOT0 (ablation) | FLOT · KITTI_o test | EPE: 0.106 -> 0.110 | Hurt |
| use FLOT0 attention-like transport cost instead of FlowNet3D mixing | FLOT0 · FlyingThings3D_o test | AS (%): 25.4 -> 33.8 | Improved |
| use FLOT0 instead of FlowNet3D | FLOT0 · KITTI_o test | AS (%): 27.6 -> 45.3 | Improved |
Closest papers by meaning
- Self-Point-Flow: Self-Supervised Scene Flow Estimation From Point Clouds With Optimal Transport and Random Walk (CVPR 2021 · similarity 0.90)
- SCOOP: Self-Supervised Correspondence and Optimization-Based Scene Flow (CVPR 2023 · similarity 0.89)
- Deep Scene Flow Learning: From 2D Images to 3D Point Clouds (TPAMI 2023 · similarity 0.89)
- LinNet: Linear Network for Efficient Point Cloud Representation Learning (NeurIPS 2024 · similarity 0.86)
- Rectified Point Flow: Generic Point Cloud Pose Estimation (NeurIPS 2025 · similarity 0.86)
- ZeroFlow: Scalable Scene Flow via Distillation (ICLR 2024 · similarity 0.86)
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.
