Research map / 3D vision / Depth and stereo
RAFT-3D: Scene Flow Using Rigid-Motion Embeddings
Zachary Teed, Jia Deng
Check 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 |
|---|---|---|---|
| represent per-pixel motion as an SE3 field and use rigid-motion embeddings with Dense-SE3 updates | RAFT-3D versus RAFT (3D flow) · FlyingThings3D (FlowNet3D split) | δ3D<.05: 36.2% -> 87.8% | Improved |
| represent motion with structured SE3 transformations rather than point-cloud flow | RAFT-3D versus FLOT · FlyingThings3D (FlowNet3D split) | δ3D<.05: 34.3% -> 83.7% | Improved |
| use RAFT-3D instead of DRISF | KITTI scene-flow estimation · KITTI | Scene Flow all: 6.31 -> 5.77 | Improved |
| increase update iterations from 1 to 16 (ablation) | RAFT-3D · FlyingThings3D (FlowNet3D split) | δ3D<.05: 56.0 -> 87.1 | Improved |
| increase update iterations from 16 to 32 (ablation) | RAFT-3D · FlyingThings3D (FlowNet3D split) | δ3D<.05: 87.1 -> 87.0 | Hurt |
| increase Dense-SE3 neighborhood radius from 8 px to 256 px (ablation) | RAFT-3D · FlyingThings3D (FlowNet3D split) | δ3D<.05: 38.7 -> 87.1 | Improved |
| expand the Dense-SE3 neighborhood from radius 256 px to the full image (ablation) | RAFT-3D · FlyingThings3D (FlowNet3D split) | δ3D<.05: 87.1 -> 83.2 | Hurt |
| add inverse-depth revisions to flow revisions (ablation) | RAFT-3D · FlyingThings3D (FlowNet3D split) | δ3D<.05: 84.6 -> 87.1 | Improved |
| add bi-Laplacian embedding smoothing (ablation) | RAFT-3D · FlyingThings3D (FlowNet3D split) | δ3D<.05: 87.1 -> 87.8 | Improved |
Later papers that compared with this method
6 later accepted papers report RAFT-3D as a baseline in their results tables; 4 of them report a better number on most of the results they share with it.
- x^2-Fusion: Cross-Modality and Cross-Dimension Flow Estimation in Event Edge Space (CVPR 2026 · better on 2 of 2 shared results)
- SEA-Flow3D: Simplified, Efficient, and Accurate Scene Flow via Spatial Vector Sampling and Multi-scale Refinement (CVPR 2026 · better on 2 of 2 shared results)
- OAMaskFlow: Occlusion-Aware Motion Mask for Scene Flow (AAAI 2025 · better on 2 of 2 shared results)
- RPEFlow: Multimodal Fusion of RGB-PointCloud-Event for Joint Optical Flow and Scene Flow Estimation (ICCV 2023 · better on 1 of 2 shared results)
- Learning Optical Expansion From Scale Matching (CVPR 2023 · better on 0 of 1 shared result)
- CamLiFlow: Bidirectional Camera-LiDAR Fusion for Joint Optical Flow and Scene Flow Estimation (CVPR 2022 · better on 2 of 2 shared results)
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
- RAFT-MSF: Self-Supervised Monocular Scene Flow Using Recurrent Optimizer (IJCV 2023 · similarity 0.89)
- What Matters for 3D Scene Flow Network (ECCV 2022 · similarity 0.87)
- OAMaskFlow: Occlusion-Aware Motion Mask for Scene Flow (AAAI 2025 · similarity 0.85)
- DeltaFlow: An Efficient Multi-frame Scene Flow Estimation Method (NeurIPS 2025 · similarity 0.84)
- World Tracing: Pixel-Aligned Geometry Beyond the Visible (NeurIPS 2026 · similarity 0.84)
- Recovering Dynamic 3D Sketches from Videos (CVPR 2025 · 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.
