ConDL: Detector-Free Dense Image Matching

TL;DR

This paper presents ConDL, a deep-learning framework for dense image matching that generates pixel-wise features without keypoints, trained on synthetic data with distortions for robust correspondence estimation.

Contribution

Introduces a detector-free, fully convolutional deep learning model trained on synthetic distorted data for robust dense image matching.

Findings

Achieves high invariance to distortions like illumination and perspective changes.

Eliminates the need for keypoint detectors in dense matching.

Demonstrates superior robustness compared to traditional methods.

Abstract

In this work, we introduce a deep-learning framework designed for estimating dense image correspondences. Our fully convolutional model generates dense feature maps for images, where each pixel is associated with a descriptor that can be matched across multiple images. Unlike previous methods, our model is trained on synthetic data that includes significant distortions, such as perspective changes, illumination variations, shadows, and specular highlights. Utilizing contrastive learning, our feature maps achieve greater invariance to these distortions, enabling robust matching. Notably, our method eliminates the need for a keypoint detector, setting it apart from many existing image-matching techniques.