Research map / 3D vision / Depth and stereo
Adaptive Surface Normal Constraint for Geometric Estimation From Monocular Images
Xiaoxiao Long, Y. Zheng, Yupeng Zheng, Beiwen Tian, Cheng Chung Lin, Lingjie Liu, Hao Zhao, Guyue Zhou, Wenping Wang
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 |
|---|---|---|---|
| add adaptive surface normal constraint instead of virtual normal (ablation) | depth model with L1 (HRNet-18) · NYUD-V2 | recovered normal mean error: 31.7 -> 22.2 | Improved |
| add ASN constraint to L1 depth loss (ablation) | HRNet-18 · NYUD-V2 | recovered normal mean error: 31.3 -> 22.2 | Improved |
| use Sobel-like or least-square normal constraint instead of ASN (ablation) | depth model with L1 · NYUD-V2 | depth rel: 0.113 -> 0.118 | Hurt |
| add geometric context adaption to area adaption (ablation) | ASN · NYUD-V2 | recovered normal mean error: 22.6 -> 22.2 | Improved |
| add area adaption to geometric context adaption (ablation) | ASN · NYUD-V2 | recovered normal mean error: 22.3 -> 22.2 | Improved |
| increase local patch size from 3 to 7 (ablation) | ASN · NYUD-V2 | recovered normal mean error: 22.5 -> 22.2 | No clear effect |
| replace HRNet-18 with HRNet-48 backbone (ablation) | ASN · NYUD-V2 | depth rel: 0.111 -> 0.101 | Improved |
| replace HRNet-48 with DPT backbone (ablation) | ASN · NYUD-V2 | depth delta1: 0.890 -> 0.908 | Improved |
| add first-order geometric context guidance to normal branch (ablation) | normal estimation (DPT) · NYUD-V2 | predicted normal mean error: 15.40 -> 15.29 | No clear effect |
| add first-order geometric context guidance to normal branch (ablation) | normal estimation (DPT) · SEVERS | predicted normal median error: 1.84 -> 1.39 | Improved |
| use second-order instead of first-order guidance (ablation) | normal estimation (DPT) · NYUD-V2 | predicted normal mean error: 15.29 -> 15.30 | No clear effect |
| add normal prediction layer (ablation) | depth model (DPT) · NYUD-V2 | recovered normal mean error: 19.9 -> 18.2 | Improved |
Closest papers by meaning
- Adaptive Surface Normal Constraint for Depth Estimation (ICCV 2021 · similarity 0.96)
- CORE: Co-planarity Regularized Monocular Geometry Estimation with Weak Supervision (ICCV 2023 · similarity 0.90)
- Enforcing Geometric Constraints of Virtual Normal for Depth Prediction (ICCV 2019 · similarity 0.90)
- Virtual Normal: Enforcing Geometric Constraints for Accurate and Robust Depth Prediction (TPAMI 2021 · similarity 0.90)
- Learning Normals of Noisy Points by Local Gradient-Aware Surface Filtering (ICCV 2025 · similarity 0.89)
- SurGe: Improved Surface Geometry in Point Maps (NeurIPS 2026 · 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.
