Research map / 3D vision / Depth and stereo
SynLF: Zero-Shot Metric Depth from Light Field Cameras via Physics-Grounded Synthesis
Zhexuan Cao, Yuduo Guo, Peisheng Ding, Zhan Shi, Hui Qiao
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 |
|---|---|---|---|
| train on PG-LF synthesis instead of HCI | VisDepth · Real-world indoor LF dataset (zero-shot) | MAE (mm): 52.95 -> 30.05 | Improved |
| train on PG-LF synthesis instead of HCI | PlaneNet · Real-world indoor LF dataset (zero-shot) | MAE (mm): 91.31 -> 53.11 | Improved |
| use VisDepth instead of PlaneNet under PG-LF supervision | PG-LF training · Real-world indoor LF dataset (zero-shot) | MAE (mm): 53.11 -> 30.05 | Improved |
| add monocular prior initialization (ablation) | VisDepth · Real-world indoor LF dataset (zero-shot) | MAE (mm): 35.75 -> 30.05 | Improved |
| add explicit visibility reasoning (ablation) | VisDepth · Real-world indoor LF dataset (zero-shot) | MAE (mm): 39.03 -> 30.05 | Improved |
| add non-Lambertian perturbation (ablation) | PG-LF synthesis · Real-world indoor LF dataset (zero-shot) | MAE (mm): 50.01 -> 43.25 | Improved |
| add depth-dependent defocus (ablation) | PG-LF synthesis · Real-world indoor LF dataset (zero-shot) | MAE (mm): 50.01 -> 46.28 | Improved |
| add both defocus and non-Lambertian perturbation (ablation) | PG-LF synthesis · Real-world indoor LF dataset (zero-shot) | MAE (mm): 50.01 -> 30.05 | Improved |
| increase GRU iterations from 6 to 24 (ablation) | VisDepth · Real-world indoor LF dataset (zero-shot) | MAE (mm): 36.39 -> 30.05 | Improved |
| scale frozen prior from DAv2-Small to DAv2-Large (ablation) | VisDepth · Real-world indoor LF dataset (zero-shot) | MAE (mm) | No clear effect |
| train on PG-LF instead of HCI, cross-dataset | SynLF · HCI 4D light field | MSE x100: 0.69 (PlaneNet, HCI-trained) vs 2.08 (SynLF) | Hurt |
| train on PG-LF instead of HCI, cross-dataset | SynLF · Inria Dense light field | BP0.07: SynLF 6.45 vs OACC-Net 2.82 | Hurt |
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.
- PlaneNet — PlaneNet: Piece-Wise Planar Reconstruction From a Single RGB Image (CVPR 2018) · better on 3 of 4 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
- Towards Multimodal Depth Estimation From Light Fields (CVPR 2022 · similarity 0.88)
- Shape Estimation from Shading, Defocus, and Correspondence Using Light-Field Angular Coherence (TPAMI 2016 · similarity 0.88)
- Accurate Depth and Normal Maps From Occlusion-Aware Focal Stack Symmetry (CVPR 2017 · similarity 0.88)
- Depth Pro: Sharp Monocular Metric Depth in Less Than a Second (ICLR 2025 · similarity 0.88)
- DepthMaster: Unified Monocular Depth Estimation for Perspective and Panoramic Images (NeurIPS 2026 · similarity 0.88)
- Repurposing Marigold for Zero-Shot Metric Depth Estimation via Defocus Blur Cues (NeurIPS 2025 · 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.
