2025 CVPR CVPR 2025

MaDCoW: Marginal Distortion Correction for Wide-Angle Photography with Arbitrary Objects

Abstract

We introduce MaDCoW, a method for correcting marginal distortion of arbitrary objects in wide-angle photography. People often use wide-angle photography to convey natural scenes--smartphones typically default to wide-angle photography--but depicting very wide-field-of-view scenes produces distorted object appearance, particularly marginal distortion in linear projections. With MaDCoW, a user annotates regions-of-interest to correct, along with straight lines. For each region, MaDCoW solves for a local-linear perspective projection and then jointly solves for a projection for the whole photograph that minimizes distortion. We show that our method can produce good results in cases where previous methods yield visible distortions.

🌉 Interdisciplinary Bridge — Computer Science and Computer Vision and Mathematics & Optimization
🧭 Keyword Pioneer — wide-angle photography
🐝 Cross-Pollinator — Artificial Intelligence, Computer Science, Computer Vision, Deep Learning, Interdisciplinary, Machine Learning, Mathematics & Optimization