CAFT：层次化视觉-语言学习提升长文本描述的视觉接地理解

View PDF HTML (experimental)

Abstract: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 Forest 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.

Comments:	Accepted to NeurIPS 2026. Project page: this https URL
Subjects:	Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
Cite as:	arXiv:2602.02977 [cs.CV]
	(or arXiv:2602.02977v3 [cs.CV] for this version)
	https://doi.org/10.48550/arXiv.2602.02977 arXiv-issued DOI via DataCite

Submission history

From: Byeongju Woo [view email]
[v1] Tue, 3 Feb 2026 01:31:55 UTC (5,423 KB)
[v2] Wed, 13 May 2026 06:25:34 UTC (5,375 KB)
[v3] Tue, 10 Nov 2026 06:37:03 UTC (5,389 KB)