Instance-level Consistent Graph With Unsupervised Human Parts for Person Re-identification
Keywords: Person re-identification, Instance-level consistency, Human parts clustering, Graph convolution network
TL;DR: We propose an Instance-level Consistent Graph (ICG) framework to address this issue, which extracts structural information by introducing graph modeling atop unsupervised human parts.
Abstract: The representation of human parts plays a crucial role in person re-identification (re-ID) by offering discriminative cues, yet it presents challenges such as misalignment, occlusion, and extreme illumination. Previous methods have primarily focused on achieving strict part-level consistency. However, individual part features change inevitably under harsh conditions, hindering consistent representation. In this article, we propose an Instance-level Consistent Graph (ICG) framework to address this issue, which extracts structural information by introducing graph modeling atop unsupervised human parts. Firstly, we introduce an attention-based foreground separation to suppress non-instance noise. Subsequently, an unsupervised clustering method is designed to segment pixel-wise human parts within the foreground, enabling fine-grained part representations. We propose a flexible structure graph that derives instance-level structure from part features, treating each part feature as a node in a graph convolutional network. In essence, ICG mitigates incompleteness through feature flow among nodes, broadening the matching condition from strict part-level consistency to robust instance-level consistency. Extensive experiments on three popular person re-ID datasets demonstrate that ICG surpasses most state-of-the-art methods, exhibiting remarkable improvements over the baseline.
Primary Area: applications to computer vision, audio, language, and other modalities
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 2543
Loading