Research map / 3D vision / Depth and stereo
LumiDepth: Stable Monocular Depth in Multi-Illumination Scenes
Anqi Cheng, Zhiyuan Yang, Tianjiao Li, Haiyue Zhu, Kezhi Mao
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 |
|---|---|---|---|
| DCPS probabilistic selection instead of best/mean/median (ablation) | LumiDepth · ReMID | AbsRel: 0.088 (Best) -> 0.085 | Improved |
| DCPS instead of Best selection (ablation) | LumiDepth · NuScenes-Night | AbsRel: 0.245 -> 0.223 | Improved |
| FaCD instead of naive latent consistency (ablation) | LumiDepth n-step · ReMID | AbsRel: 0.365 -> 0.138 | Improved |
| FaCD (LF+HF) instead of LF only (ablation) | LumiDepth n-step · ReMID | AbsRel: 0.145 -> 0.138 | Improved |
| add pseudo loss (DCPS pseudo labels) (ablation) | LumiDepth · ReMID | MeanDV: 0.035 -> 0.030 | Improved |
| add frequency consistency terms (ablation) | LumiDepth · ReMID | MeanDV: 0.042 -> 0.030 | Improved |
| add GT depth loss (ablation) | LumiDepth · ReMID | AbsRel: MeanDV 0.029 -> 0.030; AbsRel 0.092 -> 0.085; NuScenes AbsRel 0.264 -> 0.223 | Mixed |
| plug DCPS+FaCD into discriminative DAv2 (ablation) | DAv2 · ReMID | MeanDV / AbsRel: 0.051 -> 0.036; 0.119 -> 0.102 | Improved |
| naive self-training on 14K RGB-only data (no DCPS/FaCD) (ablation) | Lotus · NYUv2 | AbsRel: 0.054 -> 0.057 | Hurt |
| LumiDepth training on multi-illumination RGB data | Lotus · ReMID | AbsRel: 0.113 -> 0.085 | Improved |
| LumiDepth training | Lotus · NYUv2 | AbsRel: 0.054 -> 0.052 | Improved |
| LumiDepth training | Lotus · RoboDepth | MeanDV: 0.051 -> 0.038 | Improved |
Closest papers by meaning
- UniLumos: Fast and Unified Image and Video Relighting with Physics-Plausible Feedback (NeurIPS 2025 · similarity 0.88)
- UniRelight: Learning Joint Decomposition and Synthesis for Video Relighting (NeurIPS 2025 · similarity 0.86)
- MultiDiff: Consistent Novel View Synthesis from a Single Image (CVPR 2024 · similarity 0.86)
- SyncLight: Single-Edit Multi-View Relighting (NeurIPS 2026 · similarity 0.86)
- Luminance-Aware Statistical Quantization: Unsupervised Hierarchical Learning for Illumination Enhancement (NeurIPS 2025 · similarity 0.86)
- Generative Multiview Relighting for 3D Reconstruction under Extreme Illumination Variation (CVPR 2025 · similarity 0.85)
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.
