Research map / 3D vision / Depth and stereo
Competitive Collaboration: Joint Unsupervised Learning of Depth, Camera Motion, Optical Flow and Motion Segmentation
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 depth, camera motion, and flow using Competitive Collaboration (ablation) | CC depth model · KITTI Eigen split | AbsRel: 0.168 -> {0.148} | Helped |
| Add SSIM loss (ablation) | Basic DispNet depth model · KITTI Eigen split | AbsRel: 0.184 -> 0.168 | Helped |
| Replace DispNet with DispResNet (ablation) | CC depth model with FlowNetC · KITTI Eigen split | AbsRel: 0.148 -> 0.144 | Helped |
| Replace FlowNetC with PWC Net (ablation) | CC depth model with DispResNet · KITTI Eigen split | AbsRel: 0.144 -> {0.140} | Helped |
| Add Cityscapes training data (ablation) | CC depth model · KITTI Eigen split | AbsRel: {0.140} -> \textbf{0.139} | Helped |
| Use joint CC instead of training static and moving-region reconstructors independently (ablation) | Optical flow estimation · KITTI 2015 training set, all pixels | EPE: 13.54; 14.68 -> 7.76 | Helped |
| Use DispResNet and PWC Net instead of DispNet and FlowNetC (ablation) | CC optical flow model · KITTI 2015 training set, all pixels | EPE: 7.76 -> 6.21 | Helped |
| Fine-tune CC unsupervised with lambda_M=0.02 (ablation) | CC optical flow model · KITTI 2015 training set | EPE: 6.21 -> 5.66 | Helped |
| Combine MaskNet and flow-consensus masks (ablation) | Motion segmentation · KITTI 2015 training set, car pixels | Overall IoU: 41.64 / 51.52 -> 56.94 | Helped |
| Share the encoder between camera-motion and motion-segmentation networks (ablation) | Camera motion estimation · KITTI Odometry Sequence 09 | Absolute Trajectory Error: 0.012 $\pm$ 0.007 -> 0.017 $\pm$ 0.009 | Hurt |
| Share the encoder between camera-motion and motion-segmentation networks (ablation) | Camera motion estimation · KITTI Odometry Sequence 10 | Absolute Trajectory Error: 0.012 $\pm$ 0.008 -> 0.015 $\pm$ 0.009 | Hurt |
| Use CC instead of independent basic training (ablation) | Depth and camera-motion estimation · KITTI Eigen split | AbsRel: 0.168 -> {0.148} | Helped |
Closest papers by meaning
- Feature-Level Collaboration: Joint Unsupervised Learning of Optical Flow, Stereo Depth and Camera Motion (CVPR 2021 · similarity 0.90)
- GeoNet: Unsupervised Learning of Dense Depth, Optical Flow and Camera Pose (CVPR 2018 · similarity 0.90)
- RM-Depth: Unsupervised Learning of Recurrent Monocular Depth in Dynamic Scenes (CVPR 2022 · similarity 0.89)
- Learning Segmentation from Point Trajectories (NeurIPS 2024 · similarity 0.87)
- Unsupervised Semantic Segmentation Through Depth-Guided Feature Correlation and Sampling (CVPR 2024 · similarity 0.87)
- UnSAMFlow: Unsupervised Optical Flow Guided by Segment Anything Model (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.