Research map / 3D vision / Depth and stereo
GeoNet: Geometric Neural Network for Joint Depth and Surface Normal Estimation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 geometric depth-to-normal and normal-to-depth modules to joint prediction | VGG-16 baseline on NYU v2 · NYU v2 test set | depth RMSE: 0.626 -> 0.608 | Helped |
| Add geometric depth-to-normal and normal-to-depth modules to joint prediction | ResNet-50 depth prediction on NYU v2 · NYU v2 test set | depth RMSE: 0.584 -> 0.569 | Helped |
| Add the depth-to-normal network to the VGG-16 baseline | Surface-normal prediction on NYU v2 · NYU v2 test set | mean angle error: 19.4 -> 19.0 | Helped |
| Use a specialized least-square geometric module instead of a vanilla CNN mapping depth to normals | Depth-to-normal estimation · NYU v2 test set | mean angle error: 47.8 -> 11.5 | Helped |
| Add the residual module to the least-square depth-to-normal module (ablation) | Depth-to-normal network · NYU v2 test set | mean angle error: 11.5 -> 8.2 | Helped |
| Use vanilla CNNs to learn depth-to-normal mapping (ablation) | Depth-to-normal estimation · NYU v2 test set | mean angle error: 39.5; 39.8; 47.8 | Hurt |
| Use geometric modules to enforce consistency between depth and normals | GeoNet versus VGG-16 baseline · NYU v2 test set | depth-to-normal mean angle error, prediction consistency: 42.2 -> 34.9 | Helped |
| Use geometric modules to improve depth-derived normals against ground truth | GeoNet versus VGG-16 baseline · NYU v2 test set | depth-to-normal mean angle error against ground-truth normals: 47.8 -> 36.8 | Helped |
Closest papers by meaning
- Unsupervised Learning of Geometry From Videos With Edge-Aware Depth-Normal Consistency (AAAI 2018 · similarity 0.89)
- VPLNet: Deep Single View Normal Estimation With Vanishing Points and Lines (CVPR 2020 · similarity 0.88)
- Geometry and Learning Co-Supported Normal Estimation for Unstructured Point Cloud (CVPR 2020 · similarity 0.88)
- Just Look at the Image: Viewpoint-Specific Surface Normal Prediction for Improved Multi-View Reconstruction (CVPR 2016 · similarity 0.87)
- Deep Iterative Surface Normal Estimation (CVPR 2020 · similarity 0.87)
- Depth from a Single Image by Harmonizing Overcomplete Local Network Predictions (NeurIPS 2016 · 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.