Reasoning as Representation: Rethinking Visual Reinforcement Learning in Image Quality Assessment

Shijie Zhao#, Xuanyu Zhang#, Weiqi Li, Junlin Li, Li Zhang, Tianfan Xue, Jian Zhang
ICLR 2026 Oral
Paper | Code
We revisit the reasoning mechanism in MLLM-based IQA model and propose a CLIP-based lightweight image scorer RALI. We verifies that through RL training, MLLMs leverage their reasoning capability to convert redundant visual representations into compact, cross-domain aligned text representations. This conversion is the source of the generalization exhibited by these reasoning-based IQA models. RALI uses only about 4% of Q-Insight’s parameters and inference time, while achieving comparable accuracy.

VQ-Insight: Teaching VLMs for AI-Generated Video Quality Understanding via Progressive Visual Reinforcement Learning

Xuanyu Zhang, Weiqi Li, Shijie Zhao, Junlin Li, Li Zhang, Jian Zhang
AAAI 2026 Oral
Paper | Code
We propose VQ-Insight, a novel reasoning-style VLM framework for AIGC video quality assessment. Our approach features: (1) a progressive video quality learning scheme; (2) the design of multi-dimension scoring rewards, preference comparison rewards, and temporal modeling rewards.

Q-Insight: Understanding Image Quality via Visual Reinforcement Learning

Weiqi Li, Xuanyu Zhang, Shijie Zhao, Yabin Zhang, Junlin Li, Li Zhang and Jian Zhang
NeurIPS 2025 Spotlight
Paper | Code
We propose Q-Insight, a reinforcement learning-based model built upon group relative policy optimization (GRPO), which demonstrates strong visual reasoning capability for image quality understanding while requiring only a limited amount of rating scores and degradation labels.

FakeShield: Explainable Image Forgery Detection and Localization via Multi-modal Large Language Models

Zhipei Xu#, Xuanyu Zhang#, Runyi Li, Zecheng Tang, Qing Huang, Jian Zhang
ICLR 2025
Project | Code
We propose the explainable IFDL task and design FakeShield, a multi-modal framework capable of evaluating image authenticity, generating tampered region masks, and providing a judgment basis based on pixel-level and image-level tampering clues.

OmniGuard: Hybrid Manipulation Localization via Augmented Versatile Deep Image Watermarking

Xuanyu Zhang, Zecheng Tang, Zhipei Xu, Runyi Li, Youmin Xu, Bin Chen, Feng Gao, Jian Zhang
CVPR 2025
Paper | Code
We propose OmniGuard, a novel augmented versatile watermarking approach that integrates proactive embedding with passive, blind extraction for robust copyright protection and tamper localization.

GS-Hider: Hiding Messages into 3D Gaussian Splatting

Xuanyu Zhang, Jiarui Meng, Runyi Li, Zhipei Xu, Yongbing Zhang, Jian Zhang
NeurIPS 2024
Paper | Code
We propose the first 3DGS steganography framework GS-Hider, which can hide an entire 3D scene or an image into the original 3D scene and accurately decode it from 3D Gaussians.

EditGuard: Versatile Image Watermarking for Tamper Localization and Copyright Protection

Xuanyu Zhang, Runyi Li, Jiwen Yu, Youmin Xu, Weiqi Li, Jian Zhang
CVPR 2024
Paper | Code
We propose a versatile deep forensic watermark for AIGC editing methods, such as stable diffusion inpaint, controlnet, SDXL and etc.