Towards Precise End-to-end Weakly Supervised Object Detection Network

TL;DR

This paper introduces an end-to-end weakly supervised object detection network that jointly trains multiple instance learning and bounding-box regression, improving localization accuracy without instance-level annotations.

Contribution

It proposes a unified network with shared backbone and guided attention, enabling end-to-end training and surpassing previous two-phase methods.

Findings

Achieves state-of-the-art performance on public datasets.

Effectively extracts implicit location information via guided attention.

Joint training reduces local minima issues in object detection.

Abstract

It is challenging for weakly supervised object detection network to precisely predict the positions of the objects, since there are no instance-level category annotations. Most existing methods tend to solve this problem by using a two-phase learning procedure, i.e., multiple instance learning detector followed by a fully supervised learning detector with bounding-box regression. Based on our observation, this procedure may lead to local minima for some object categories. In this paper, we propose to jointly train the two phases in an end-to-end manner to tackle this problem. Specifically, we design a single network with both multiple instance learning and bounding-box regression branches that share the same backbone. Meanwhile, a guided attention module using classification loss is added to the backbone for effectively extracting the implicit location information in the features. Experimental results on public datasets show that our method achieves state-of-the-art performance.