STNet: Selective Tuning of Convolutional Networks for Object Localization

TL;DR

STNet introduces a novel feedback-based approach combining bottom-up and top-down processing in convolutional networks to improve object localization and generate attention maps, surpassing previous state-of-the-art methods.

Contribution

The paper proposes STNet, a new model that integrates feedback mechanisms into CNNs for enhanced object localization and attention map generation.

Findings

STNet outperforms existing methods on ImageNet localization.

It effectively generates attention-driven class hypothesis maps.

The approach demonstrates the importance of feedback in visual hierarchy modeling.

Abstract

Visual attention modeling has recently gained momentum in developing visual hierarchies provided by Convolutional Neural Networks. Despite recent successes of feedforward processing on the abstraction of concepts form raw images, the inherent nature of feedback processing has remained computationally controversial. Inspired by the computational models of covert visual attention, we propose the Selective Tuning of Convolutional Networks (STNet). It is composed of both streams of Bottom-Up and Top-Down information processing to selectively tune the visual representation of Convolutional networks. We experimentally evaluate the performance of STNet for the weakly-supervised localization task on the ImageNet benchmark dataset. We demonstrate that STNet not only successfully surpasses the state-of-the-art results but also generates attention-driven class hypothesis maps.