OpenVision 3: A Family of Unified Visual Encoder for Both Understanding and Generation

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Zhang, Letian, Ren, Sucheng, Liu, Yanqing, Li, Xianhang, Wang, Zeyu, Zhou, Yuyin, Yao, Huaxiu, Zheng, Zeyu, Nie, Weili, Liu, Guilin, Yu, Zhiding, Xie, Cihang
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912964169170944
author Zhang, Letian
Ren, Sucheng
Liu, Yanqing
Li, Xianhang
Wang, Zeyu
Zhou, Yuyin
Yao, Huaxiu
Zheng, Zeyu
Nie, Weili
Liu, Guilin
Yu, Zhiding
Xie, Cihang
author_facet Zhang, Letian
Ren, Sucheng
Liu, Yanqing
Li, Xianhang
Wang, Zeyu
Zhou, Yuyin
Yao, Huaxiu
Zheng, Zeyu
Nie, Weili
Liu, Guilin
Yu, Zhiding
Xie, Cihang
contents This paper presents a family of advanced vision encoder, named OpenVision 3, that learns a single, unified visual representation that can serve both image understanding and image generation. Our core architecture is simple: we feed VAE-compressed image latents to a ViT encoder and train its output to support two complementary roles. First, the encoder output is passed to the ViT-VAE decoder to reconstruct the original image, encouraging the representation to capture generative structure. Second, the same representation is optimized with contrastive learning and image-captioning objectives, strengthening semantic features. By jointly optimizing reconstruction- and semantics-driven signals in a shared latent space, the encoder learns representations that synergize and generalize well across both regimes. We validate this unified design through extensive downstream evaluations with the encoder frozen. For generation, we test it under the RAE framework: ours substantially surpasses the standard CLIP-based encoder (e.g., gFID: 1.87 vs. 2.54 on ImageNet). For multimodal understanding, we plug the encoder into the LLaVA-1.5 and LLaVA-NeXT framework: it performs comparably with a standard CLIP vision encoder (e.g., 63.3 vs. 61.2 on SeedBench, and 59.2 vs. 58.1 on GQA). We provide empirical evidence that generation and understanding are mutually beneficial in our architecture, while further underscoring the critical role of the VAE latent space. We hope this work can spur future research on unified modeling.
format Preprint
id arxiv_https___arxiv_org_abs_2601_15369
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle OpenVision 3: A Family of Unified Visual Encoder for Both Understanding and Generation
Zhang, Letian
Ren, Sucheng
Liu, Yanqing
Li, Xianhang
Wang, Zeyu
Zhou, Yuyin
Yao, Huaxiu
Zheng, Zeyu
Nie, Weili
Liu, Guilin
Yu, Zhiding
Xie, Cihang
Image and Video Processing
Artificial Intelligence
This paper presents a family of advanced vision encoder, named OpenVision 3, that learns a single, unified visual representation that can serve both image understanding and image generation. Our core architecture is simple: we feed VAE-compressed image latents to a ViT encoder and train its output to support two complementary roles. First, the encoder output is passed to the ViT-VAE decoder to reconstruct the original image, encouraging the representation to capture generative structure. Second, the same representation is optimized with contrastive learning and image-captioning objectives, strengthening semantic features. By jointly optimizing reconstruction- and semantics-driven signals in a shared latent space, the encoder learns representations that synergize and generalize well across both regimes. We validate this unified design through extensive downstream evaluations with the encoder frozen. For generation, we test it under the RAE framework: ours substantially surpasses the standard CLIP-based encoder (e.g., gFID: 1.87 vs. 2.54 on ImageNet). For multimodal understanding, we plug the encoder into the LLaVA-1.5 and LLaVA-NeXT framework: it performs comparably with a standard CLIP vision encoder (e.g., 63.3 vs. 61.2 on SeedBench, and 59.2 vs. 58.1 on GQA). We provide empirical evidence that generation and understanding are mutually beneficial in our architecture, while further underscoring the critical role of the VAE latent space. We hope this work can spur future research on unified modeling.
title OpenVision 3: A Family of Unified Visual Encoder for Both Understanding and Generation
topic Image and Video Processing
Artificial Intelligence
url https://arxiv.org/abs/2601.15369