Convolutional STN for Weakly Supervised Object Localization

TL;DR

This paper introduces a convolutional multi-scale spatial localization network that enhances weakly supervised object localization by providing more accurate localization, outperforming traditional CNN-based methods on standard datasets.

Contribution

The paper proposes a novel convolutional, multi-scale spatial localization network specifically designed for weakly supervised object localization tasks.

Findings

Achieves competitive localization performance on CUB-200-2011 dataset.

Outperforms baseline methods on ImageNet dataset.

Provides a more accurate localization mechanism compared to standard CNN activation map approaches.

Abstract

Weakly supervised object localization is a challenging task in which the object of interest should be localized while learning its appearance. State-of-the-art methods recycle the architecture of a standard CNN by using the activation maps of the last layer for localizing the object. While this approach is simple and works relatively well, object localization relies on different features than classification, thus, a specialized localization mechanism is required during training to improve performance. In this paper, we propose a convolutional, multi-scale spatial localization network that provides accurate localization for the object of interest. Experimental results on CUB-200-2011 and ImageNet datasets show that our proposed approach provides competitive performance for weakly supervised localization.