Coordinate Representation

Coordinate representation in computer vision focuses on developing robust and efficient methods for encoding spatial information, particularly to address challenges like ambiguous feature matching in images and videos. Recent research emphasizes geometrically invariant coordinate systems, such as barycentric coordinates, which offer improved resilience to transformations and repeated patterns compared to standard Cartesian coordinates. Prominent approaches involve neural networks that learn probabilistic coordinate fields or anchor-to-barycentric mappings, often incorporating confidence measures to enhance reliability. These advancements improve performance in various applications, including image registration, 3D scene stylization, and feature correspondence tasks, leading to more accurate and robust visual processing systems.

Papers

June 7, 2023

Learning Probabilistic Coordinate Fields for Robust Correspondences
Weiyue Zhao, Hao Lu, Xinyi Ye, Zhiguo Cao, Xin Li
Robust Correspondence Barycentric Coordinate Correspondence Based Method Coordinate Representation

June 5, 2023

A2B: Anchor to Barycentric Coordinate for Robust Correspondence
Weiyue Zhao, Hao Lu, Zhiguo Cao, Xin Li
Visual Information Anchor Learning Correspondence Image Correspondence Noisy Correspondence Barycentric Coordinate Sparse Correspondence Coordinate Representation

April 5, 2022

Unified Implicit Neural Stylization
Zhiwen Fan, Yifan Jiang, Peihao Wang, Xinyu Gong, Dejia Xu, Zhangyang Wang
Neural Radiance Field Implicit Representation Implicit Surface Neural Stylization Coordinate Representation

Coordinate Representation

Papers

Learning Probabilistic Coordinate Fields for Robust Correspondences

A2B: Anchor to Barycentric Coordinate for Robust Correspondence

Unified Implicit Neural Stylization