Research map / 3D vision / Face and shape
Marr Revisited: 2D-3D Alignment via Surface Normal Prediction
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 |
|---|---|---|---|
| Use a combination of low-, mid-, and high-level VGG-16 layers rather than only low-level layers (ablation) | Surface normal skip network · NYUv2 test | Mean angular error: 44.4 -> 19.8 | Helped |
| Add layer 7 to the five-layer set {1₂, 2₂, 3₃, 4₃, 5₃} (ablation) | Surface normal skip network · NYUv2 test | Mean angular error: 20.9 -> 19.8 | Helped |
| Apply Manhattan-world rectification to the predicted normals | Surface normal skip network · NYUv2 test, global scene layout | Mean angular error: 19.8 -> 23.9 | Hurt |
| Use Geom scoring instead of dot-product scoring for predicted-normal nearest-neighbor retrieval | Nearest-neighbor CAD retrieval · NYUD2 test, chairs with >50% valid depth | Area under pose accuracy curve: 0.21 -> 0.31 | Helped |
| Add pool-5 appearance features to predicted-normal retrieval | Nearest-neighbor CAD retrieval · NYUD2 test, chairs with >50% valid depth | Area under pose accuracy curve: 0.21 -> 0.25; 0.31 -> 0.29 | Mixed |
| Add pool-5 appearance features to depth-normal retrieval | Nearest-neighbor CAD retrieval · NYUD2 test, chairs with >50% valid depth | Area under pose accuracy curve: 0.41 -> 0.36; 0.44 -> 0.38 | Hurt |
Closest papers by meaning
- Just Look at the Image: Viewpoint-Specific Surface Normal Prediction for Improved Multi-View Reconstruction (CVPR 2016 · similarity 0.89)
- Scan2CAD: Learning CAD Model Alignment in RGB-D Scans (CVPR 2019 · similarity 0.88)
- SURGE: Surface Regularized Geometry Estimation from a Single Image (NeurIPS 2016 · similarity 0.88)
- ROCA: Robust CAD Model Retrieval and Alignment From a Single Image (CVPR 2022 · similarity 0.87)
- PMNI: Pose-free Multi-view Normal Integration for Reflective and Textureless Surface Reconstruction (CVPR 2025 · similarity 0.87)
- Front2Back: Single View 3D Shape Reconstruction via Front to Back Prediction (CVPR 2020 · 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.