Research map / Image generation / Image restoration
Rethinking Generic Camera Models for Deep Single Image Camera Calibration to Recover Rotation and Fisheye Distortion
Nobuhiko Wakai, Satoshi Sato, Yasunori Ishii, Takayoshi Yamashita
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 harmonic non-grid bearing loss (HNGBL) joint weights (ablation) | Ours (DenseNet-161 calibration) · StreetLearn | REPE (pixel): 12.65 -> 7.39 | Improved |
| add HNGBL joint weights (ablation) | Ours · SP360 | REPE (pixel): 12.57 -> 7.39 | Improved |
| add HNGBL joint weights (ablation) | Ours · StreetLearn | tilt error (deg): 7.23 -> 4.13 | Improved |
| add HNGBL joint weights (ablation) | Ours · StreetLearn | roll error (deg): 7.73 -> 5.21 | Improved |
| add HNGBL joint weights (ablation) | Ours · SP360 | focal length error (mm): 0.49 -> 0.39 | Improved |
| add HNGBL joint weights (ablation) | Ours · StreetLearn | PSNR: 27.41 -> 29.01 | Improved |
| add HNGBL joint weights (ablation) | Ours · SP360 | SSIM: 0.801 -> 0.835 | Improved |
| add HNGBL joint weights (ablation) | Ours · StreetLearn orthogonal projection | PSNR: 23.97 -> 23.70 | Hurt |
| add HNGBL joint weights (ablation) | Ours · SP360 equidistance projection | PSNR: 28.53 -> 29.28 | Improved |
| replace Wakai's equisolid-only model and equal loss weights with the generic camera model and HNGBL | Wakai's method · StreetLearn | REPE (pixel): 30.02 -> 7.39 | Improved |
| use proposed generic camera model instead of equisolid projection | Wakai's method · StreetLearn | PSNR: 21.57 -> 29.01 | Improved |
| use orthogonal instead of equidistance projection images | Ours · StreetLearn trigonometric models | PSNR: 30.10 -> 23.70 | Hurt |
Closest papers by meaning
- Learning to Calibrate Straight Lines for Fisheye Image Rectification (CVPR 2019 · similarity 0.88)
- Deep Single Image Camera Calibration by Heatmap Regression to Recover Fisheye Images Under Manhattan World Assumption (CVPR 2024 · similarity 0.88)
- Deep Single Image Camera Calibration With Radial Distortion (CVPR 2019 · similarity 0.87)
- Extending Foundational Monocular Depth Estimators to Fisheye Cameras with Calibration Tokens (ICCV 2025 · similarity 0.87)
- Self-Calibrating Neural Radiance Fields (ICCV 2021 · similarity 0.86)
- Why Having 10,000 Parameters in Your Camera Model Is Better Than Twelve (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.
