Research map / 3D vision / Depth and stereo
Seeing and Seeing Through the Glass: Real and Synthetic Data for Multi-Layer Depth Estimation
Hongyu Wen, Yiming Zuo, Venkat Subramanian, Patrick Chen, Jia Deng
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 |
|---|---|---|---|
| fine-tune on the synthetic LayeredDepth-Syn data | Metric3D V2 · LayeredDepth real benchmark (Layer 1) | quadruplet accuracy: 55.14 -> 75.20 | Improved |
| fine-tune on the synthetic data | Metric3D V2 · LayeredDepth real benchmark (Layer 1) | pair accuracy: 80.31 -> 89.53 | Improved |
| fine-tune on the synthetic data | Metric3D V2 · LayeredDepth real benchmark (Layer 1) | triplet accuracy: 65.43 -> 81.71 | Improved |
| use Layer Index Concatenation instead of Multi-head output | NeWCRFs multi-layer baseline · LayeredDepth real benchmark (all tuples) | quadruplet accuracy: 25.97 -> 26.00 | Improved |
| use Recurrent instead of Multi-head output | NeWCRFs multi-layer baseline · LayeredDepth real benchmark (all tuples) | quadruplet accuracy: 25.97 -> 24.64 | Hurt |
| use Recurrent instead of Multi-head output | NeWCRFs multi-layer baseline · LayeredDepth real benchmark (Layer 7) | quadruplet accuracy: 33.10 -> 27.59 | Hurt |
| use Recurrent instead of Multi-head output | NeWCRFs multi-layer baseline · LayeredDepth real benchmark (Layer 1) | quadruplet accuracy: 24.21 -> 26.12 | Improved |
| replace NeWCRFs backbone with Depth Anything V2 backbone (multi-head) | Multi-head baseline · LayeredDepth real benchmark (all tuples) | quadruplet accuracy: 25.97 -> 61.68 | Improved |
| replace NeWCRFs backbone with Depth Anything V2 backbone (multi-head) | Multi-head baseline · LayeredDepth synthetic validation (Layer 1) | AbsRel (x100): 17.97 -> 8.24 | Improved |
| evaluate with Adapted Layer instead of First Layer strategy | Metric3D V2 (metric) · LayeredDepth synthetic validation (transparent regions) | AbsRel: 0.31 -> 0.16 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- Depth Pro — Depth Pro: Sharp Monocular Metric Depth in Less Than a Second (ICLR 2025) · better on 2 of 2 shared results
- MoGe — MoGe: Unlocking Accurate Monocular Geometry Estimation for Open-Domain Images with Optimal Training Supervision (CVPR 2025) · better on 1 of 1 shared result
- GeoWizard — GeoWizard: Unleashing the Diffusion Priors for 3D Geometry Estimation from a Single Image (ECCV 2024) · better on 1 of 1 shared result
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
- SeeGroup: Multi-Layer Depth Estimation of Transparent Surfaces via Self-Determined Grouping (CVPR 2026 · similarity 0.91)
- PolarDepth: Monocular Transparent Object Depth from Polar-Physics Priors (ICML 2026 · similarity 0.88)
- One Scene, Two Depths: Probing Geometric Ambiguity in Monocular Foundation Models (ECCV 2026 · similarity 0.88)
- SeeClear: Reliable Transparent Object Depth Estimation via Generative Opacification (ECCV 2026 · similarity 0.88)
- SharpDepth: Sharpening Metric Depth Predictions Using Diffusion Distillation (CVPR 2025 · similarity 0.87)
- From Transparent to Opaque: Rethinking Neural Implicit Surfaces with $\alpha$-NeuS (NeurIPS 2024 · similarity 0.86)
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.
