Saved in:
Bibliographic Details
Main Authors: Woo, Byeongju, Wang, Zilin, Pak, Byeonghyun, Mo, Sangwoo, Yu, Stella X.
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2602.02977
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918499125821440
author Woo, Byeongju
Wang, Zilin
Pak, Byeonghyun
Mo, Sangwoo
Yu, Stella X.
author_facet Woo, Byeongju
Wang, Zilin
Pak, Byeonghyun
Mo, Sangwoo
Yu, Stella X.
contents Vision-language models such as CLIP often struggle to faithfully understand long, detail-rich captions, relying on dominant scene cues while overlooking fine-grained visual evidence. We propose a hierarchical vision-language learning principle for understanding scenes as part-to-whole compositions: before forming a whole-scene representation, a model should uncover what semantic parts appear where in the image. To this end, we propose CAFT (Cross-domain Alignment of Forests and Trees), a vision-language model that jointly learns local text-region alignment at intermediate representations and global image-text alignment at the final representation. Exploiting the organization of long captions, where local descriptions often correspond to scene parts, CAFT employs a fine-to-coarse image encoder and a part-whole text encoder to discover localized part semantics and progressively compose them into a global image-text representation. Trained on 30M image-text pairs, CAFT achieves state-of-the-art performance on six long-text retrieval benchmarks and exhibits strong scaling behavior. Experiments show that CAFT learns fine-grained representations that localize textual semantics in image regions without explicit region-level supervision.
format Preprint
id arxiv_https___arxiv_org_abs_2602_02977
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Aligning Forest and Trees in Images & Long Captions for Visually Grounded Understanding
Woo, Byeongju
Wang, Zilin
Pak, Byeonghyun
Mo, Sangwoo
Yu, Stella X.
Computer Vision and Pattern Recognition
Artificial Intelligence
Machine Learning
Vision-language models such as CLIP often struggle to faithfully understand long, detail-rich captions, relying on dominant scene cues while overlooking fine-grained visual evidence. We propose a hierarchical vision-language learning principle for understanding scenes as part-to-whole compositions: before forming a whole-scene representation, a model should uncover what semantic parts appear where in the image. To this end, we propose CAFT (Cross-domain Alignment of Forests and Trees), a vision-language model that jointly learns local text-region alignment at intermediate representations and global image-text alignment at the final representation. Exploiting the organization of long captions, where local descriptions often correspond to scene parts, CAFT employs a fine-to-coarse image encoder and a part-whole text encoder to discover localized part semantics and progressively compose them into a global image-text representation. Trained on 30M image-text pairs, CAFT achieves state-of-the-art performance on six long-text retrieval benchmarks and exhibits strong scaling behavior. Experiments show that CAFT learns fine-grained representations that localize textual semantics in image regions without explicit region-level supervision.
title Aligning Forest and Trees in Images & Long Captions for Visually Grounded Understanding
topic Computer Vision and Pattern Recognition
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2602.02977