Research map / 3D vision / Depth and stereo
LightDepth: Single-View Depth Self-Supervision from Illumination Decline
Javier Rodríguez-Puigvert, Víctor M. Batlle, J.M.M. Montiel, Ruben Martinez-Cantin, Pascal Fua, Juan D. Tardós, Javier Civera
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 |
|---|---|---|---|
| replace depth ground-truth supervision with single-view light-decline self-supervision (U-Net) | U-Net ResNet18 · Synthetic colon | depth MedAE (mm): 2.99 -> 2.47 | Improved |
| replace depth ground-truth supervision with light self-supervision | U-Net ResNet18 · Synthetic colon | depth RMSE (mm): 6.38 -> 8.60 | Hurt |
| replace depth ground-truth supervision with light self-supervision | U-Net ResNet18 · Synthetic colon | normal MAE (degrees): 25.1 -> 15.2 | Improved |
| add test-time refinement (20 steps per image) (ablation) | LightDepth U-Net · C3VD | depth MAE (mm): 4.37 -> 3.72 | Improved |
| light self-supervision instead of multi-view self-supervision (Monodepth2) | C3VD · C3VD | depth MAE (mm): 14.27 -> 4.37 | Improved |
| train on target domain with light self-supervision instead of synthetic-to-real transfer | XDCycleGAN baseline · C3VD | depth MAE (mm): 17.16 -> 4.37 | Improved |
| light self-supervision instead of depth GT when training on synthetic and testing on C3VD | U-Net · Synthetic to C3VD | depth MAE (mm): 9.44 -> 5.09 | Improved |
| add smoothness loss L_s (ablation) | LightDepth U-Net · C3VD | depth MAE (mm): 6.05 -> 4.95 | Improved |
| add specular loss L_sp (ablation) | LightDepth U-Net · C3VD | depth MAE (mm): 4.95 -> 4.37 | Improved |
| add specular loss L_sp (ablation) | LightDepth U-Net · C3VD | color MAE of rendered image: 0.0690 -> 0.0657 | Mixed |
| light self-supervision versus depth GT with DPT-Hybrid on C3VD | DPT-Hybrid · C3VD | depth MAE (mm): 3.22 (GT) vs 3.94 (Light) | Hurt |
| compute normals with in-house six-neighbourhood instead of U-Net regressor (ablation) | Normals from depth · C3VD | normal MAE (degrees): 16.24 -> 1.32 | Improved |
Closest papers by meaning
- DeLightMono: Enhancing Self-Supervised Monocular Depth Estimation in Endoscopy by Decoupling Uneven Illumination (AAAI 2026 · similarity 0.89)
- Leveraging Near-Field Lighting for Monocular Depth Estimation from Endoscopy Videos (ECCV 2024 · similarity 0.88)
- Depth Any Endoscopy: Towards Self-Supervised Generalizable Depth Estimation in Monocular Endoscopy (CVPR 2026 · similarity 0.87)
- Unsupervised Photometric-Consistent Depth Estimation from Endoscopic Monocular Video (AAAI 2025 · similarity 0.84)
- Self-supervised Monocular Depth Estimation Robust to Reflective Surface Leveraged by Triplet Mining (ICLR 2025 · similarity 0.83)
- Intrinsic Image Decomposition for Robust Self-supervised Monocular Depth Estimation on Reflective Surfaces (AAAI 2025 · similarity 0.83)
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.
