ePrints@IIScePrints@IISc Home | About | Browse | Latest Additions | Advanced Search | Contact | Help

G-PReDICT: Generalizable Person Re-ID using Domain Invariant Contrastive Techniquesg±

Jambigi, C and Masud, U and Chakraborty, A (2022) G-PReDICT: Generalizable Person Re-ID using Domain Invariant Contrastive Techniquesg±. In: 13th Indian Conference on Computer Vision, Graphics, and Image Processing,, 8 - 10 Dec 2022, Gandhinagar.

[img] PDF
ACM_int_con_pro_ser_2022.pdf - Published Version
Restricted to Registered users only

Download (5MB) | Request a copy
Official URL: https://doi.org/10.1145/3571600.3571655


Learning identity-Aware, domain-invariant representations is crucial in solving domain generalizable person ReID (DG-ReID). Existing methods commonly use augmentation techniques either in feature space by mixing instance and batch normalization layers or in pixel space by adversarially generating pseudo domains. However, neither of these techniques guarantee identity preservation. Apart from increasing training data diversity, the augmented positive pairs also encode rich semantic relations which have not been fully explored. To address the above issues, we propose a novel framework for Generalizable Person Re-identification using Domain Invariant Contrastive Techniques (G-PReDICT). Specifically, we use simple yet effective perturbation strategies to hallucinate positive samples across domains by realistically modelling domain variations while preserving the target identities. We harness rich sample-sample relations between the hallucinated positive-negative pairs to learn domain-invariant representations using supervised contrastive learning. We also use a domain independent auxiliary task, i.e. attribute prediction to learn robust representations and introduce attribute annotations for two large scale public benchmarks i.e. CUHK-03 and MSMT17. Extensive experiments on standard benchmarks demonstrate the effectiveness of the proposed method.

Item Type: Conference Paper
Publication: ACM International Conference Proceeding Series
Publisher: Association for Computing Machinery
Additional Information: The copyright for this article belongs to Association for Computing Machinery.
Keywords: Deep learning; Domain Generalization Computer vision; Person Re-Identification Association for Computing Machinery
Department/Centre: Division of Interdisciplinary Sciences > Computational and Data Sciences
Date Deposited: 30 Jun 2023 04:56
Last Modified: 30 Jun 2023 04:56
URI: https://eprints.iisc.ac.in/id/eprint/82140

Actions (login required)

View Item View Item