Weakly-supervised Discovery of Visual Pattern Configurations

TL;DR

This paper introduces a weakly-supervised method that automatically discovers visual pattern configurations to improve object detection, achieving state-of-the-art results on PASCAL VOC by addressing mislocalizations and enhancing training data quality.

Contribution

It formulates the discovery of discriminative visual configurations as a constrained submodular optimization problem, advancing weakly-supervised object detection techniques.

Findings

Achieved state-of-the-art weakly-supervised detection on PASCAL VOC.

Effectively remedies mislocalizations using discovered configurations.

Identifies informative positive and negative training examples.

Abstract

The increasing prominence of weakly labeled data nurtures a growing demand for object detection methods that can cope with minimal supervision. We propose an approach that automatically identifies discriminative configurations of visual patterns that are characteristic of a given object class. We formulate the problem as a constrained submodular optimization problem and demonstrate the benefits of the discovered configurations in remedying mislocalizations and finding informative positive and negative training examples. Together, these lead to state-of-the-art weakly-supervised detection results on the challenging PASCAL VOC dataset.