Negative Prototypes Guided Contrastive Learning for WSOD

TL;DR

This paper introduces a novel contrastive learning framework guided by negative prototypes for weakly supervised object detection, effectively utilizing inter-image relationships and a global feature bank to improve detection accuracy.

Contribution

The paper proposes a Negative Prototypes Guided Contrastive learning architecture that leverages negative prototypes and a global feature bank to enhance WSOD performance.

Findings

Achieves state-of-the-art results on VOC07 and VOC12 datasets.

Effectively distinguishes between categories using negative prototypes.

Improves feature representation through contrastive learning with reliable pseudo labels.

Abstract

Weakly Supervised Object Detection (WSOD) with only image-level annotation has recently attracted wide attention. Many existing methods ignore the inter-image relationship of instances which share similar characteristics while can certainly be determined not to belong to the same category. Therefore, in order to make full use of the weak label, we propose the Negative Prototypes Guided Contrastive learning (NPGC) architecture. Firstly, we define Negative Prototype as the proposal with the highest confidence score misclassified for the category that does not appear in the label. Unlike other methods that only utilize category positive feature, we construct an online updated global feature bank to store both positive prototypes and negative prototypes. Meanwhile, we propose a pseudo label sampling module to mine reliable instances and discard the easily misclassified instances based on the feature similarity with corresponding prototypes in global feature bank. Finally, we follow the contrastive learning paradigm to optimize the proposal's feature representation by attracting same class samples closer and pushing different class samples away in the embedding space. Extensive experiments have been conducted on VOC07, VOC12 datasets, which shows that our proposed method achieves the state-of-the-art performance.