English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Weakly supervised text-based person re-identification (TPRe-ID) seeks toretrieve images of a target person using textual descriptions, without relyingon identity annotations and is more challenging and practical. The primarychallenge is the intra-class differences, encompassing intra-modal featurevariations and cross-modal semantic gaps. Prior works have focused oninstance-level samples and ignored prototypical features of each person whichare intrinsic and invariant. Toward this, we propose a Cross-Modal PrototypicalContrastive Learning (CPCL) method. In practice, the CPCL introduces the CLIPmodel to weakly supervised TPRe-ID for the first time, mapping visual andtextual instances into a shared latent space. Subsequently, the proposedPrototypical Multi-modal Memory (PMM) module captures associations betweenheterogeneous modalities of image-text pairs belonging to the same personthrough the Hybrid Cross-modal Matching (HCM) module in a many-to-many mappingfashion. Moreover, the Outlier Pseudo Label Mining (OPLM) module furtherdistinguishes valuable outlier samples from each modality, enhancing thecreation of more reliable clusters by mining implicit relationships betweenimage-text pairs. Experimental results demonstrate that our proposed CPCLattains state-of-the-art performance on all three public datasets, with asignificant improvement of 11.58%, 8.77% and 5.25% in Rank@1 accuracy onCUHK-PEDES, ICFG-PEDES and RSTPReid datasets, respectively. The code isavailable at https://github.com/codeGallery24/CPCL.

CPCL: Cross-Modal Prototypical Contrastive Learning for Weakly Supervised Text-based Person Re-Identification