Research map / 3D vision / Depth and stereo
Self-Supervised Learning With Geometric Constraints in Monocular Video: Connecting Flow, Depth, and Camera
Yuhua Chen, Cordelia Schmid, Cristian Sminchisescu
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 adaptive photometric loss (ablation) | standard photometric baseline · KITTI Eigen depth | Abs Rel: 0.156 -> 0.144 | Improved |
| add multi-view structure consistency loss (ablation) | L_apc · KITTI Eigen depth | Abs Rel: 0.144 -> 0.138 | Improved |
| add epipolar constraint loss (ablation) | L_apc + L_mvs · KITTI Eigen depth | Abs Rel: 0.138 -> 0.135 | Improved |
| add online refinement (ablation) | L_apc + L_mvs + L_e · KITTI Eigen depth | Abs Rel: 0.135 -> 0.099 | Improved |
| add online refinement (ablation) | L_apc + L_mvs · KITTI Eigen depth | Abs Rel: 0.138 -> 0.103 | Improved |
| add output-space regularizer to finetuning (ablation) | standard parameter finetuning · KITTI Eigen depth (refinement) | Abs Rel | Improved |
| use output finetuning instead of parameter finetuning (ablation) | online refinement · KITTI Eigen depth (refinement) | speed: 2 seconds vs 40 seconds for 50 iterations | No clear effect |
| combine PFT and OFT (ablation) | PFT alone · KITTI Eigen depth (refinement) | Abs Rel | Improved |
| add epipolar constraint loss to flow training (ablation) | L_apc + L_mvs · KITTI 2015 flow (train set) | EPE All: 12.20 -> 8.95 | Improved |
| add adaptive photometric and multi-view losses (ablation) | flow baseline · KITTI 2015 flow (train set) | EPE All: 12.28 -> 12.20 | No clear effect |
| add online refinement (ablation) | flow with all losses · KITTI 2015 flow (train set) | EPE All: 8.95 -> 8.35 | Improved |
| predict intrinsics instead of using ground-truth calibration (ablation) | GLNet depth (KITTI-trained) · KITTI Eigen depth | Abs Rel: 0.099 -> 0.100 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- GeoNet — GeoNet: Geometric Neural Network for Joint Depth and Surface Normal Estimation (CVPR 2018) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- GeoNet: Unsupervised Learning of Dense Depth, Optical Flow and Camera Pose (CVPR 2018 · similarity 0.89)
- Depth From Videos in the Wild: Unsupervised Monocular Depth Learning From Unknown Cameras (ICCV 2019 · similarity 0.89)
- The Temporal Opportunist: Self-Supervised Multi-Frame Monocular Depth (CVPR 2021 · similarity 0.89)
- Revisit Self-supervision with Local Structure-from-Motion (ECCV 2024 · similarity 0.88)
- Geometry Guided Convolutional Neural Networks for Self-Supervised Video Representation Learning (CVPR 2018 · similarity 0.88)
- Mining Supervision for Dynamic Regions in Self-Supervised Monocular Depth Estimation (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.
