Research map / 3D vision / Depth and stereo
UnOS: Unified Unsupervised Optical-Flow and Stereo-Depth Estimation by Watching Videos
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 |
|---|---|---|---|
| jointly train optical flow and stereo depth with rigid-region flow consistency | UnOS · KITTI 2012 | EPE: 1.26 -> 1.04 | Helped |
| jointly train optical flow and stereo depth with rigid-region flow consistency | UnOS · KITTI 2015 train all | EPE: 7.76 -> 5.58 | Helped |
| add consecutive-frame information and ego-motion training (ablation) | UnOS stereo depth · KITTI 2015 train | Abs Rel: 0.060 -> 0.052 | Helped |
| add RDVO and rigid-aware flow consistency (ablation) | UnOS stereo depth · KITTI 2015 train | Abs Rel: 0.052 -> 0.049 | Helped |
| add RDVO refinement to MotionNet pose (ablation) | UnOS visual odometry · KITTI sequence 09 | Absolute trajectory error: 0.023 ± 0.010 -> 0.013 ± 0.006 | Helped |
| add final joint training with flow consistency after RDVO (ablation) | UnOS visual odometry · KITTI sequence 09 | Absolute trajectory error: 0.013 ± 0.006 -> 0.012 ± 0.006 | Helped |
| use stereo input and joint depth-flow learning | UnOS · KITTI 2015 train | D1-all: 5.140% -> 5.943% | Hurt |
| use rigid flow from ego-motion instead of learned full optical flow (ablation) | UnOS optical flow · KITTI 2015 train all | EPE: 7.88 -> 11.9 | Hurt |
| add RDVO refinement to ego-motion rigid flow (ablation) | UnOS rigid flow · KITTI 2015 train all | EPE: 11.9 -> 10.69 | Helped |
| use StereoNet alone without temporal information (ablation) | UnOS stereo depth · KITTI 2015 train | Abs Rel: 0.060 -> 0.052 | Hurt |
| jointly estimate motion and stereo depth rather than using StereoNet alone (ablation) | UnOS stereo depth · KITTI 2015 train | Sq Rel: 0.833 -> 0.593 | Helped |
Closest papers by meaning
- Feature-Level Collaboration: Joint Unsupervised Learning of Optical Flow, Stereo Depth and Camera Motion (CVPR 2021 · similarity 0.91)
- GeoNet: Unsupervised Learning of Dense Depth, Optical Flow and Camera Pose (CVPR 2018 · similarity 0.89)
- Depth Prediction without the Sensors: Leveraging Structure for Unsupervised Learning from Monocular Videos (AAAI 2019 · similarity 0.88)
- EffiScene: Efficient Per-Pixel Rigidity Inference for Unsupervised Joint Learning of Optical Flow, Depth, Camera Pose and Motion Segmentation (CVPR 2021 · similarity 0.88)
- UnSAMFlow: Unsupervised Optical Flow Guided by Segment Anything Model (CVPR 2024 · similarity 0.86)
- Depth2Pose: A Pose-Based Benchmark for Monocular Depth Estimation without Ground-Truth Depth (NeurIPS 2026 · 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.