Research map / 3D vision / Depth and stereo
MaDCoW: Marginal Distortion Correction for Wide-Angle Photography with Arbitrary Objects
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Introduces MaDCoW, a user-guided wide-angle reprojection method that optimizes a linear perspective projection per annotated region of interest and then solves a joint whole-image warp preserving lines and conformality.
Problem
Marginal distortion of objects in very wide-field-of-view linear perspective photographs; existing methods only correct specific content types (faces, lines, building facades) and leave distortion elsewhere.
Approach
Applies the Direct View Condition per user-annotated region by fitting a linear perspective projection (optical axis plus similarity transform) to each region's boundary under stereographic projection, then blends these per-region projections into a full-image warp with line, conformality, smoothness and DVC terms.
Key results
- In a 44-participant user study, viewers preferred MaDCoW results over Carroll et al. 81% of the time overall and judged objects on margins more natural 84% of the time.
- Per-image user preference for MaDCoW over line-only Carroll et al. ranged from 59% to 95% overall across seven comparisons.
- Method takes about two minutes per example versus about one minute for the Carroll et al. implementation.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add MaDCoW region annotations and projection optimization instead of line annotations only | Wide-angle photography · User study, overall image comparisons | preference for MaDCoW: 81% | Helped |
| Add MaDCoW region annotations and projection optimization instead of line annotations only | Wide-angle photography · User study, marginal objects | preference for natural-looking objects: 84% | Helped |
| Add single-object projection optimization rather than using the object-center principal point without optimization (ablation) | MaDCoW · Single-object optimization examples | distortion | Helped |
| Optimize the principal point as well as the affine transformation (ablation) | MaDCoW · Single-object optimization examples | distortion | Helped |
Surprising findings
- Adding more line annotations does not fix shape distortion of buildings with the Carroll et al. method, since line losses compete with other terms.
- Adding straight-line constraints to a people-only image did not help and was omitted.
Limitations
- Requires user annotation of distorted regions and line constraints; optimal masks may need refinement.
- Dense ROIs or features crossing multiple ROIs produce artifacts or bent lines.
- Large regions spanning wide fields of view retain some marginal distortion; optimization is too slow for real-time use.
Experimental setup
Metrics: user preference rate
Baselines: Carroll et al. (line-annotation warp) Shih et al. (face stereographic correction) stereographic projection
Builds on: Carroll et al. conformal warp Direct View Condition Zorin and Barr curvilinear projection
Closest papers by meaning
- Practical Wide-Angle Portraits Correction With Deep Structured Models (CVPR 2021 · similarity 0.84)
- PRaDA: Projective Radial Distortion Averaging (CVPR 2025 · similarity 0.82)
- Splatter-360: Generalizable 360 Gaussian Splatting for Wide-baseline Panoramic Images (CVPR 2025 · similarity 0.82)
- Deep Single Image Camera Calibration by Heatmap Regression to Recover Fisheye Images Under Manhattan World Assumption (CVPR 2024 · similarity 0.82)
- Wild-GS: Real-Time Novel View Synthesis from Unconstrained Photo Collections (NeurIPS 2024 · similarity 0.81)
- Extreme Rotation Estimation in the Wild (CVPR 2025 · similarity 0.81)
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.