Research map / 3D vision / Depth and stereo
Forget About the LiDAR: Self-Supervised Depth Estimators with MED Probability Volumes
Juan Luis GonzalezBello, Munchurl Kim
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 |
|---|---|---|---|
| regress disparity in exponential/MED probability volume rather than linear space (ablation) | FAL-netB33 · KITTI improved Eigen test | abs rel: 0.109 -> 0.079 | Improved |
| increase parameters from 6.6M to 17M (ablation) | FAL-net w/o MOM · KITTI improved Eigen test | abs rel: 0.087 -> 0.079 | Improved |
| increase parameters from 17M to 26M (ablation) | FAL-net w/o MOM · KITTI improved Eigen test | abs rel: 0.079 -> 0.080 | No clear effect |
| use 49 instead of 33 disparity levels (ablation) | FAL-net w/o MOM · KITTI improved Eigen test | abs rel: 0.079 -> 0.076 | No clear effect |
| train concurrently on K+CS instead of K only (ablation) | FAL-netB49 w/o MOM · KITTI improved Eigen test | abs rel: 0.076 -> 0.074 | Improved |
| CS pre-training then K re-training instead of K only (ablation) | FAL-netB49 w/o MOM · KITTI improved Eigen test | abs rel: 0.076 -> 0.085 | Hurt |
| train 20 extra epochs without MOM (ablation) | FAL-netB33 w/o MOM · KITTI improved Eigen test | abs rel: 0.079 -> 0.081 | Hurt |
| train MOM model from scratch for 50 epochs instead of fine-tuning (ablation) | FAL-netB33 · KITTI improved Eigen test | abs rel: 0.078 vs 0.076 | No clear effect |
| fine-tune with Mirror Occlusion Module (MOM) (ablation) | FAL-netB33 · KITTI improved Eigen test | abs rel: 0.079 -> 0.076 | Improved |
| fine-tune with MOM (ablation) | FAL-netB49 K+CS · KITTI improved Eigen test | abs rel: 0.074 -> 0.071 | Improved |
| add multi-scale post-processing | FAL-netB49 · KITTI original Eigen test | abs rel: 0.097 -> 0.093 | Improved |
| add CityScapes to training (K+CS) | FAL-netB49 with PP · KITTI original Eigen test | abs rel: 0.093 -> 0.088 | Improved |
Closest papers by meaning
- Digging Into Self-Supervised Monocular Depth Estimation (ICCV 2019 · similarity 0.91)
- Revealing the Reciprocal Relations Between Self-Supervised Stereo and Monocular Depth Estimation (ICCV 2021 · similarity 0.91)
- Self-Supervised Monocular Trained Depth Estimation Using Self-Attention and Discrete Disparity Volume (CVPR 2020 · similarity 0.91)
- Self-Supervised Monocular Depth Hints (ICCV 2019 · similarity 0.91)
- Mind The Edge: Refining Depth Edges in Sparsely-Supervised Monocular Depth Estimation (CVPR 2024 · similarity 0.89)
- HELVIPAD: A Real-World Dataset for Omnidirectional Stereo Depth Estimation (CVPR 2025 · similarity 0.88)
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.
