Improving Text-based Person Search via Part-level Cross-modal Correspondence

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Park, Jicheol, Jeong, Boseung, Kim, Dongwon, Kwak, Suha
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866912174658551808
author Park, Jicheol
Jeong, Boseung
Kim, Dongwon
Kwak, Suha
author_facet Park, Jicheol
Jeong, Boseung
Kim, Dongwon
Kwak, Suha
contents Text-based person search is the task of finding person images that are the most relevant to the natural language text description given as query. The main challenge of this task is a large gap between the target images and text queries, which makes it difficult to establish correspondence and distinguish subtle differences across people. To address this challenge, we introduce an efficient encoder-decoder model that extracts coarse-to-fine embedding vectors which are semantically aligned across the two modalities without supervision for the alignment. There is another challenge of learning to capture fine-grained information with only person IDs as supervision, where similar body parts of different individuals are considered different due to the lack of part-level supervision. To tackle this, we propose a novel ranking loss, dubbed commonality-based margin ranking loss, which quantifies the degree of commonality of each body part and reflects it during the learning of fine-grained body part details. As a consequence, it enables our method to achieve the best records on three public benchmarks.
format Preprint
id arxiv_https___arxiv_org_abs_2501_00318
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Improving Text-based Person Search via Part-level Cross-modal Correspondence
Park, Jicheol
Jeong, Boseung
Kim, Dongwon
Kwak, Suha
Computer Vision and Pattern Recognition
Machine Learning
Text-based person search is the task of finding person images that are the most relevant to the natural language text description given as query. The main challenge of this task is a large gap between the target images and text queries, which makes it difficult to establish correspondence and distinguish subtle differences across people. To address this challenge, we introduce an efficient encoder-decoder model that extracts coarse-to-fine embedding vectors which are semantically aligned across the two modalities without supervision for the alignment. There is another challenge of learning to capture fine-grained information with only person IDs as supervision, where similar body parts of different individuals are considered different due to the lack of part-level supervision. To tackle this, we propose a novel ranking loss, dubbed commonality-based margin ranking loss, which quantifies the degree of commonality of each body part and reflects it during the learning of fine-grained body part details. As a consequence, it enables our method to achieve the best records on three public benchmarks.
title Improving Text-based Person Search via Part-level Cross-modal Correspondence
topic Computer Vision and Pattern Recognition
Machine Learning
url https://arxiv.org/abs/2501.00318