Automatic discovery of discriminative parts as a quadratic assignment problem

TL;DR

This paper introduces a quadratic assignment approach to automatically discover discriminative parts for image classification, achieving state-of-the-art results on public datasets.

Contribution

It formulates part learning as a quadratic assignment problem, enabling automatic and optimal correspondence between image regions and parts.

Findings

Achieves state-of-the-art accuracy on Willow actions dataset.

Achieves state-of-the-art accuracy on MIT 67 scenes dataset.

Analyzes and compares different assignment strategies.

Abstract

Part-based image classification consists in representing categories by small sets of discriminative parts upon which a representation of the images is built. This paper addresses the question of how to automatically learn such parts from a set of labeled training images. The training of parts is cast as a quadratic assignment problem in which optimal correspondences between image regions and parts are automatically learned. The paper analyses different assignment strategies and thoroughly evaluates them on two public datasets: Willow actions and MIT 67 scenes. State-of-the art results are obtained on these datasets.