Convolutional neural network architecture for geometric matching

TL;DR

This paper introduces a trainable convolutional neural network architecture for geometric image matching, capable of estimating transformations and matching images at both instance and category levels, with strong generalization and state-of-the-art performance.

Contribution

A novel CNN architecture for geometric matching that is end-to-end trainable, trained on synthetic data, and effective for both instance and category-level matching.

Findings

Achieves state-of-the-art results on Proposal Flow dataset.

Can be trained solely on synthetic imagery without manual annotations.

Generalizes well to unseen images and different matching tasks.

Abstract

We address the problem of determining correspondences between two images in agreement with a geometric model such as an affine or thin-plate spline transformation, and estimating its parameters. The contributions of this work are three-fold. First, we propose a convolutional neural network architecture for geometric matching. The architecture is based on three main components that mimic the standard steps of feature extraction, matching and simultaneous inlier detection and model parameter estimation, while being trainable end-to-end. Second, we demonstrate that the network parameters can be trained from synthetically generated imagery without the need for manual annotation and that our matching layer significantly increases generalization capabilities to never seen before images. Finally, we show that the same model can perform both instance-level and category-level matching giving state-of-the-art results on the challenging Proposal Flow dataset.