Research map / 3D vision / Depth and stereo
Perspective Fields for Single Image Camera Calibration
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 |
|---|---|---|---|
| Predict per-pixel Perspective Fields instead of deriving them from global camera parameters | PerspectiveNet · Stanford2D3D, Crop | Up mean error (°): 4.49 -> 2.21 | Helped |
| Predict per-pixel Perspective Fields instead of deriving them from global camera parameters | PerspectiveNet · TartanAir, Crop | Latitude mean error (°): 9.22 -> 5.73 | Helped |
| Train on cropped object images using teacher-generated pseudo labels | PerspectiveNet on Objectron · Objectron, isolated | Up mean error (°): 17.24 -> 4.45 | Helped |
| Train on cropped object images using teacher-generated pseudo labels | PerspectiveNet on Objectron · Objectron, isolated | Latitude mean error (°): 84.61 -> 9.65 | Mixed |
| Remove Perspective Field loss L_pers (ablation) | ParamNet camera calibration · GSV uncentered principal point | Pitch mean error (°): 2.60 -> 2.88 | Hurt |
| Predict principal-point shift rather than assume the point is centered (ablation) | ParamNet camera calibration · GSV uncentered principal point | Pitch mean error (°): 4.23 -> 2.60 | Helped |
| Train without RandomResizedCrop so training images have centered principal points (ablation) | PerspectiveNet · Stanford2D3D, Crop | Up mean error (°): 2.21 -> 3.07 | Hurt |
| Train without RandomResizedCrop so training images have centered principal points (ablation) | PerspectiveNet · Stanford2D3D, None | Up mean error (°): 2.18 -> 1.83 | Helped |
| Initialize camera optimization with ParamNet predictions | Camera parameter optimization · GSV uncentered principal point | Latitude median error (°): 2.17 -> 1.91 | Helped |
| Measure perspective mismatch with combined Up-vector and Latitude discrepancy | Perspective Fields compositing metric · Human perspective consistency ratings | Pearson correlation (median): 0.71 -> 0.87 | Helped |
Closest papers by meaning
- A Perceptual Measure for Deep Single Image Camera Calibration (CVPR 2018 · similarity 0.86)
- 3D Reconstruction with Generalizable Neural Fields using Scene Priors (ICLR 2024 · similarity 0.83)
- CRAYM: Neural Field Optimization via Camera RAY Matching (NeurIPS 2024 · similarity 0.83)
- Photography Perspective Composition: Towards Aesthetic Perspective Recommendation (NeurIPS 2025 · similarity 0.82)
- Cameras as Relative Positional Encoding (NeurIPS 2025 · similarity 0.82)
- ProvNeRF: Modeling per Point Provenance in NeRFs as a Stochastic Field (NeurIPS 2024 · similarity 0.82)
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.