FineViT: Progressively Unlocking Fine-Grained Perception with Dense Recaptions

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Zhao, Peisen, Zhang, Xiaopeng, Xu, Mingxing, Sun, Ruoyu, Du, Zewei, Wang, Dunzheng, Zheng, Guanghao, Xu, Haohang, Zhang, Zhibo, Zhang, Yuhang, Ai, Yi, Liu, Lin, Tian, Qi
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912972418318336
author Zhao, Peisen
Zhang, Xiaopeng
Xu, Mingxing
Sun, Ruoyu
Du, Zewei
Wang, Dunzheng
Zheng, Guanghao
Xu, Haohang
Zhang, Zhibo
Zhang, Yuhang
Ai, Yi
Liu, Lin
Tian, Qi
author_facet Zhao, Peisen
Zhang, Xiaopeng
Xu, Mingxing
Sun, Ruoyu
Du, Zewei
Wang, Dunzheng
Zheng, Guanghao
Xu, Haohang
Zhang, Zhibo
Zhang, Yuhang
Ai, Yi
Liu, Lin
Tian, Qi
contents While Multimodal Large Language Models (MLLMs) have experienced rapid advancements, their visual encoders frequently remain a performance bottleneck. Conventional CLIP-based encoders struggle with dense spatial tasks due to the loss of visual details caused by low-resolution pretraining and the reliance on noisy, coarse web-crawled image-text pairs. To overcome these limitations, we introduce FineViT, a novel vision encoder specifically designed to unlock fine-grained perception. By replacing coarse web data with dense recaptions, we systematically mitigate information loss through a progressive training paradigm.: first, the encoder is trained from scratch at a high native resolution on billions of global recaptioned image-text pairs, establishing a robust, detail rich semantic foundation. Subsequently, we further enhance its local perception through LLM alignment, utilizing our curated FineCap-450M dataset that comprises over $450$ million high quality local captions. Extensive experiments validate the effectiveness of the progressive strategy. FineViT achieves state-of-the-art zero-shot recognition and retrieval performance, especially in long-context retrieval, and consistently outperforms multimodal visual encoders such as SigLIP2 and Qwen-ViT when integrated into MLLMs. We hope FineViT could serve as a powerful new baseline for fine-grained visual perception.
format Preprint
id arxiv_https___arxiv_org_abs_2603_17326
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle FineViT: Progressively Unlocking Fine-Grained Perception with Dense Recaptions
Zhao, Peisen
Zhang, Xiaopeng
Xu, Mingxing
Sun, Ruoyu
Du, Zewei
Wang, Dunzheng
Zheng, Guanghao
Xu, Haohang
Zhang, Zhibo
Zhang, Yuhang
Ai, Yi
Liu, Lin
Tian, Qi
Computer Vision and Pattern Recognition
While Multimodal Large Language Models (MLLMs) have experienced rapid advancements, their visual encoders frequently remain a performance bottleneck. Conventional CLIP-based encoders struggle with dense spatial tasks due to the loss of visual details caused by low-resolution pretraining and the reliance on noisy, coarse web-crawled image-text pairs. To overcome these limitations, we introduce FineViT, a novel vision encoder specifically designed to unlock fine-grained perception. By replacing coarse web data with dense recaptions, we systematically mitigate information loss through a progressive training paradigm.: first, the encoder is trained from scratch at a high native resolution on billions of global recaptioned image-text pairs, establishing a robust, detail rich semantic foundation. Subsequently, we further enhance its local perception through LLM alignment, utilizing our curated FineCap-450M dataset that comprises over $450$ million high quality local captions. Extensive experiments validate the effectiveness of the progressive strategy. FineViT achieves state-of-the-art zero-shot recognition and retrieval performance, especially in long-context retrieval, and consistently outperforms multimodal visual encoders such as SigLIP2 and Qwen-ViT when integrated into MLLMs. We hope FineViT could serve as a powerful new baseline for fine-grained visual perception.
title FineViT: Progressively Unlocking Fine-Grained Perception with Dense Recaptions
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2603.17326