Research map / 3D vision / Depth and stereo
Bounding Boxes, Segmentations and Object Coordinates: How Important Is Recognition for 3D Scene Flow Estimation in Autonomous Driving Scenarios?
Aseem Behl, Omid Hosseini Jafari, Siva Karthik Mustikovela, Hassan Abu Alhaija, Carsten Rother, Andreas Geiger
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 2D bounding box recognition input (ablation) | OSF · KITTI 2015 scene flow (validation) | SF error fg: 24.12 -> 17.11 | Improved |
| add 2D instance segmentation input instead of bounding boxes (ablation) | ISF-BBox · KITTI 2015 scene flow (validation) | SF error fg: 17.11 -> 14.60 | Improved |
| add instance segmentation input (ablation) | OSF · KITTI 2015 scene flow (validation) | SF error fg: 24.12 -> 14.60 | Improved |
| add instance segmentation input (ablation) | OSF · KITTI 2015 scene flow (validation) | Fl error fg: 20.56 -> 10.78 | Improved |
| add instance segmentation input (ablation) | OSF · KITTI 2015 scene flow (validation) | D1 error fg: 8.86 -> 7.97 | Improved |
| add instance segmentation input (ablation) | OSF · KITTI 2015 scene flow (validation) | SF error bg: 7.38 -> 7.74 | Hurt |
| add 3D object coordinates on top of segmentation (ablation) | ISF-SegMask · KITTI 2015 scene flow (validation) | SF error bg+fg: 8.79 -> 8.80 | No clear effect |
| add 3D object coordinates on top of segmentation (ablation) | ISF-SegMask · KITTI 2015 scene flow (validation) | SF error fg: 14.60 -> 14.66 | No clear effect |
| use CNN-based disparity input (DispNetC and MC-CNN-acrt) (ablation) | ISF-SegMask · KITTI 2015 scene flow (validation) | SF error bg+fg: 8.79 -> 7.46 | Improved |
| use CNN-based disparity input (ablation) | ISF-SegMask · KITTI 2015 scene flow (validation) | D1 error fg: 7.97 -> 3.94 | Improved |
| add RGB-XYZ input to the MNC instance network instead of RGB only (ablation) | MNC · KITTI validation set of [5] | instance IoU-50% accuracy: 78% -> 83% | Improved |
| use CNN instead of random forest for object coordinates (ablation) | object coordinate predictor · KITTI cars | average Euclidean error (m): 2.89 -> 0.6 | Improved |
Closest papers by meaning
- Weakly Supervised Learning of Rigid 3D Scene Flow (CVPR 2021 · similarity 0.90)
- SeFlow: A Self-Supervised Scene Flow Method in Autonomous Driving (ECCV 2024 · similarity 0.89)
- Zero-Shot Monocular Scene Flow Estimation in the Wild (CVPR 2025 · similarity 0.87)
- ICP-Flow: LiDAR Scene Flow Estimation with ICP (CVPR 2024 · similarity 0.87)
- DifFlow3D: Toward Robust Uncertainty-Aware Scene Flow Estimation with Iterative Diffusion-Based Refinement (CVPR 2024 · similarity 0.87)
- VoteFlow: Enforcing Local Rigidity in Self-Supervised Scene Flow (CVPR 2025 · 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.
