Research map / 3D vision / Depth and stereo
DeMoN: Depth and Motion Network for Learning Monocular Stereo
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 joint two-image depth and motion estimation with correspondence-based flow, rather than single-image depth | DeMoN · SUN3D bootstrap | L1-inv: 0.080 -> 0.012 | Helped |
| use correspondence-based flow and iterative depth/motion estimation instead of direct naïve image-pair regression | DeMoN · SUN3D bootstrap | L1-inv: 0.079 -> 0.012 | Helped |
| add scale-invariant gradient loss (ablation) | DeMoN bootstrap · SUN3D | sc-inv: 0.211 -> 0.159 | Helped |
| add scale-invariant gradient loss (ablation) | DeMoN bootstrap · SUN3D | L1-inv: 0.040 -> 0.057 | Hurt |
| add scale-invariant gradient loss (ablation) | DeMoN bootstrap · SUN3D | rot: 3.127 -> 4.585 | Hurt |
| add surface-normal prediction and loss (ablation) | DeMoN bootstrap · SUN3D | L1-inv: 0.040 -> 0.037 | Helped |
| add optical-flow loss to normal prediction and loss (ablation) | DeMoN bootstrap · SUN3D | rot: 2.570 -> 2.359 | Helped |
| add flow confidence as an input to egomotion estimation (ablation) | DeMoN bootstrap · SUN3D | rot: 2.830 -> 2.479 | Helped |
| add flow confidence as an input (ablation) | DeMoN bootstrap · SUN3D | EPE: 0.027 -> 0.027 | No clear effect |
| add flow confidence as an input (ablation) | DeMoN bootstrap · SUN3D | L1-rel: 0.26 -> 0.28 | Hurt |
| use DeMoN rather than the strongest listed classic baseline | two-frame depth and motion estimation · Sun3D | L1-inv: 0.029 -> 0.019 | Helped |
| use DeMoN rather than the strongest listed classic baseline | two-frame depth and motion estimation · MVS | rot: 4.834 -> 5.156 | Hurt |
Closest papers by meaning
- Depth Prediction without the Sensors: Leveraging Structure for Unsupervised Learning from Monocular Videos (AAAI 2019 · similarity 0.87)
- RM-Depth: Unsupervised Learning of Recurrent Monocular Depth in Dynamic Scenes (CVPR 2022 · similarity 0.87)
- DeepV2D: Video to Depth with Differentiable Structure from Motion (ICLR 2020 · similarity 0.87)
- Feature-Level Collaboration: Joint Unsupervised Learning of Optical Flow, Stereo Depth and Camera Motion (CVPR 2021 · similarity 0.87)
- Unsupervised Scale-consistent Depth and Ego-motion Learning from Monocular Video (NeurIPS 2019 · similarity 0.87)
- Competitive Collaboration: Joint Unsupervised Learning of Depth, Camera Motion, Optical Flow and Motion Segmentation (CVPR 2019 · 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.