A Solution to Co-occurrence Bias: Attributes Disentanglement via Mutual Information Minimization for Pedestrian Attribute Recognition

TL;DR

This paper introduces a method to reduce co-occurrence bias in pedestrian attribute recognition by disentangling attributes through mutual information minimization, leading to more robust and generalizable models.

Contribution

It proposes a novel attribute-disentangled feature learning approach that minimizes mutual information to decouple attributes, improving generalization across diverse scenes.

Findings

Achieves state-of-the-art performance on PETAzs and RAPzs datasets.

Substantially improves baseline accuracy and robustness.

Effectively reduces co-occurrence bias in attribute recognition.

Abstract

Recent studies on pedestrian attribute recognition progress with either explicit or implicit modeling of the co-occurrence among attributes. Considering that this known a prior is highly variable and unforeseeable regarding the specific scenarios, we show that current methods can actually suffer in generalizing such fitted attributes interdependencies onto scenes or identities off the dataset distribution, resulting in the underlined bias of attributes co-occurrence. To render models robust in realistic scenes, we propose the attributes-disentangled feature learning to ensure the recognition of an attribute not inferring on the existence of others, and which is sequentially formulated as a problem of mutual information minimization. Rooting from it, practical strategies are devised to efficiently decouple attributes, which substantially improve the baseline and establish state-of-the-art performance on realistic datasets like PETAzs and RAPzs. Code is released on https://github.com/SDret/A-Solution-to-Co-occurence-Bias-in-Pedestrian-Attribute-Recognition.