Research map / 3D vision / 3D generation
Weakly Supervised Learning of Rigid 3D Scene Flow
Check 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 |
|---|---|---|---|
| Use rigid-object weak supervision rather than full-supervised stereo-trained flow estimation | Scene flow estimation on lidarKITTI (w/o ground) · lidarKITTI (w/o ground) | EPE3D [m]: 0.390 -> 0.094 | Helped |
| Apply test-time optimization to ego-motion only | Ours weakly supervised pipeline · lidarKITTI (w/o ground) | EPE3D [m]: 0.150 -> 0.110 | Helped |
| Optimize all rigid-body transformations at test time instead of ego-motion only | Ours weakly supervised pipeline · lidarKITTI (w/o ground) | EPE3D [m]: 0.110 -> 0.094 | Helped |
| Add ego-motion loss (ablation) | Full training objective · lidarKITTI (with ground) | EPE [m]: 0.721 -> 0.134 | Helped |
| Add Chamfer distance loss (ablation) | Full training objective · lidarKITTI (with ground) | EPE [m]: 0.363 -> 0.134 | Helped |
| Add rigidity loss (ablation) | Training objective with ego-motion and Chamfer losses · lidarKITTI (with ground) | EPE [m]: 0.136 -> 0.134 | Helped |
| Initialize the weakly supervised model with the FT3D fully supervised backbone (ablation) | Ours weakly supervised pipeline · lidarKITTI (w/o ground) | EPE3D [m]: 0.150 -> 0.102 | Helped |
| Initialize the weakly supervised model with the FT3D fully supervised backbone (ablation) | Ours weakly supervised pipeline · lidarKITTI (with ground) | EPE3D [m]: 0.133 -> 0.091 | Helped |
| Use one joint multi-task network instead of task-specific networks (ablation) | Background segmentation and ego-motion estimation · semanticKITTI (w/o ground) | RRE [°]: 0.245 -> 0.201 | Helped |
| Use one joint multi-task network instead of task-specific networks (ablation) | Background segmentation and ego-motion estimation · semanticKITTI (with ground) | RRE [°]: 0.307 -> 0.249 | Helped |
| Use the joint model instead of the task-specific segmentation network (ablation) | Background segmentation · semanticKITTI | precision FG | Hurt |
| Apply test-time optimization to all rigid bodies | Ours weakly supervised pipeline · stereoKITTI (with ground) | EPE3D [m]: 0.136 -> 0.068 | Helped |
Closest papers by meaning
- VoteFlow: Enforcing Local Rigidity in Self-Supervised Scene Flow (CVPR 2025 · similarity 0.88)
- H-Flow: Self-supervised Human Scene Flow via Physics-inspired Joint Multi-modal Learning (NeurIPS 2026 · similarity 0.87)
- ZeroFlow: Scalable Scene Flow via Distillation (ICLR 2024 · similarity 0.87)
- Zero-Shot Monocular Scene Flow Estimation in the Wild (CVPR 2025 · similarity 0.87)
- 3DSFLabelling: Boosting 3D Scene Flow Estimation by Pseudo Auto-labelling (CVPR 2024 · similarity 0.87)
- ICP-Flow: LiDAR Scene Flow Estimation with ICP (CVPR 2024 · 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.