Research map / 3D vision / Depth and stereo
UprightNet: Geometry-Aware Camera Orientation Estimation From Single Images
Wenqi Xian, Zhengqi Li, Matthew Fisher, Jonathan Eisenmann, Eli Shechtman, Noah Snavely
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 predicted weights to orientation estimation (ablation) | UprightNet · InteriorNet test | mean angular error: 1.83 -> 1.17 | Improved |
| add camera orientation loss (end-to-end optimization) (ablation) | UprightNet · InteriorNet test | mean angular error: 2.71 -> 1.17 | Improved |
| add camera-frame surface supervision (ablation) | UprightNet · InteriorNet test | mean angular error: 1.70 -> 1.17 | Improved |
| add upright-frame surface supervision (ablation) | UprightNet · InteriorNet test | mean angular error: 1.82 -> 1.17 | Improved |
| use all three of n, t, b instead of normal only (ablation) | UprightNet · InteriorNet test | mean angular error: 1.81 -> 1.17 | Improved |
| use all three of n, t, b instead of t only (ablation) | UprightNet · InteriorNet test | mean angular error: 2.11 -> 1.17 | Improved |
| add t to n and b (ablation) | UprightNet · InteriorNet test | mean angular error: 1.17 -> 1.17 | Improved |
| add b to n and t (ablation) | UprightNet · InteriorNet test | mean angular error: 1.52 -> 1.17 | Improved |
| use UprightNet instead of Hold-Geoffroy et al. baseline | Hold-Geoffroy et al. · InteriorNet test | mean angular error: 1.76 -> 1.17 | Improved |
| use UprightNet instead of Hold-Geoffroy et al. baseline | Hold-Geoffroy et al. · ScanNet test | mean angular error: 4.55 -> 3.74 | Improved |
| use UprightNet roll angle versus best baseline | Hold-Geoffroy et al. · ScanNet test | mean roll error: 2.15 -> 2.12 | Improved |
| use UprightNet instead of DeepHorizon on unseen dataset | DeepHorizon · SUN360 test (trained on ScanNet) | mean angular error: 9.53 -> 7.81 | Improved |
Closest papers by meaning
- Upright-Net: Learning Upright Orientation for 3D Point Cloud (CVPR 2022 · similarity 0.90)
- Wide-Baseline Relative Camera Pose Estimation With Directional Learning (CVPR 2021 · similarity 0.88)
- RotationNet: Joint Object Categorization and Pose Estimation Using Multiviews From Unsupervised Viewpoints (CVPR 2018 · similarity 0.88)
- Orient Anything: Learning Robust Object Orientation Estimation from Rendering 3D Models (ICML 2025 · similarity 0.86)
- Orient Anything V2: Unifying Orientation and Rotation Understanding (NeurIPS 2025 · similarity 0.86)
- LEAP: Liberate Sparse-View 3D Modeling from Camera Poses (ICLR 2024 · similarity 0.85)
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.
