HGCLIP: Exploring Vision-Language Models with Graph Representations for Hierarchical Understanding

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Xia, Peng, Yu, Xingtong, Hu, Ming, Ju, Lie, Wang, Zhiyong, Duan, Peibo, Ge, Zongyuan
Natura: Preprint
Pubblicazione: 2023
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866929609800417280
author Xia, Peng
Yu, Xingtong
Hu, Ming
Ju, Lie
Wang, Zhiyong
Duan, Peibo
Ge, Zongyuan
author_facet Xia, Peng
Yu, Xingtong
Hu, Ming
Ju, Lie
Wang, Zhiyong
Duan, Peibo
Ge, Zongyuan
contents Object categories are typically organized into a multi-granularity taxonomic hierarchy. When classifying categories at different hierarchy levels, traditional uni-modal approaches focus primarily on image features, revealing limitations in complex scenarios. Recent studies integrating Vision-Language Models (VLMs) with class hierarchies have shown promise, yet they fall short of fully exploiting the hierarchical relationships. These efforts are constrained by their inability to perform effectively across varied granularity of categories. To tackle this issue, we propose a novel framework (HGCLIP) that effectively combines CLIP with a deeper exploitation of the Hierarchical class structure via Graph representation learning. We explore constructing the class hierarchy into a graph, with its nodes representing the textual or image features of each category. After passing through a graph encoder, the textual features incorporate hierarchical structure information, while the image features emphasize class-aware features derived from prototypes through the attention mechanism. Our approach demonstrates significant improvements on 11 diverse visual recognition benchmarks. Our codes are fully available at https://github.com/richard-peng-xia/HGCLIP.
format Preprint
id arxiv_https___arxiv_org_abs_2311_14064
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle HGCLIP: Exploring Vision-Language Models with Graph Representations for Hierarchical Understanding
Xia, Peng
Yu, Xingtong
Hu, Ming
Ju, Lie
Wang, Zhiyong
Duan, Peibo
Ge, Zongyuan
Computer Vision and Pattern Recognition
Object categories are typically organized into a multi-granularity taxonomic hierarchy. When classifying categories at different hierarchy levels, traditional uni-modal approaches focus primarily on image features, revealing limitations in complex scenarios. Recent studies integrating Vision-Language Models (VLMs) with class hierarchies have shown promise, yet they fall short of fully exploiting the hierarchical relationships. These efforts are constrained by their inability to perform effectively across varied granularity of categories. To tackle this issue, we propose a novel framework (HGCLIP) that effectively combines CLIP with a deeper exploitation of the Hierarchical class structure via Graph representation learning. We explore constructing the class hierarchy into a graph, with its nodes representing the textual or image features of each category. After passing through a graph encoder, the textual features incorporate hierarchical structure information, while the image features emphasize class-aware features derived from prototypes through the attention mechanism. Our approach demonstrates significant improvements on 11 diverse visual recognition benchmarks. Our codes are fully available at https://github.com/richard-peng-xia/HGCLIP.
title HGCLIP: Exploring Vision-Language Models with Graph Representations for Hierarchical Understanding
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2311.14064