Research map / 3D vision / Depth and stereo
Virtual Normal: Enforcing Geometric Constraints for Accurate and Robust Depth Prediction
Wei Yin, Yifan Liu, Chunhua Shen
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 virtual normal loss to WCEL (ablation) | WCEL baseline · NYUD-V2 | Abs-Rel: 0.1427 -> 0.1337 | Improved |
| add L1 loss to WCEL (ablation) | WCEL baseline · NYUD-V2 | Abs-Rel: 0.1427 -> 0.1429 | No clear effect |
| add surface normal loss (local) to WCEL (ablation) | WCEL baseline · NYUD-V2 | Abs-Rel: 0.1427 -> 0.1406 | Improved |
| add pair-wise geometric loss to WCEL (ablation) | WCEL baseline · NYUD-V2 | Abs-Rel: 0.1427 -> 0.1380 | Improved |
| add virtual normal loss on top of WCEL+PL (ablation) | WCEL+PL · NYUD-V2 | Abs-Rel: 0.1380 -> 0.1341 | Improved |
| increase virtual normal sample groups from 0K to 20K (ablation) | WCEL+VNL · NYUD-V2 | Abs-Rel: 5.6% | Improved |
| use multi-curriculum learning instead of uniform sampling (ablation) | DiverseDepth training · KITTI zero-shot | Abs-Rel: 17.9 -> 12.6 | Improved |
| reverse curriculum (hard to easy) instead of uniform sampling (ablation) | DiverseDepth training · KITTI zero-shot | Abs-Rel: 17.9 -> 15.8 | Improved |
| use easy-to-hard curriculum instead of reverse curriculum (ablation) | DiverseDepth training · KITTI zero-shot | Abs-Rel: 15.8 -> 12.6 | Improved |
| add SSIL to VNL (ablation) | affine-invariant depth model · KITTI zero-shot | Abs-Rel: 21.0 -> 12.6 | Improved |
| add VNL to SSIL (ablation) | affine-invariant depth model · KITTI zero-shot | Abs-Rel: 16.3 -> 12.6 | Improved |
| replace MSE loss with VNL+SSIL (ablation) | affine-invariant depth model · KITTI zero-shot | Abs-Rel: 36.0 -> 12.6 | Improved |
Closest papers by meaning
- Enforcing Geometric Constraints of Virtual Normal for Depth Prediction (ICCV 2019 · similarity 0.97)
- Towards Accurate Reconstruction of 3D Scene Shape From A Single Monocular Image (TPAMI 2022 · similarity 0.92)
- Robust Geometry-Preserving Depth Estimation Using Differentiable Rendering (ICCV 2023 · similarity 0.91)
- Metric3D v2: A Versatile Monocular Geometric Foundation Model for Zero-Shot Metric Depth and Surface Normal Estimation (TPAMI 2024 · similarity 0.91)
- Depth2Pose: A Pose-Based Benchmark for Monocular Depth Estimation without Ground-Truth Depth (NeurIPS 2026 · similarity 0.89)
- A Simple yet Universal Framework for Depth Completion (NeurIPS 2024 · 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.
