A Unified Agentic Framework for Evaluating Conditional Image Generation
Jifang Wang, Xue Yang, Longyue Wang, Zhenran Xu, Yiyu Wang, Yaowei Wang, Weihua Luo, Kaifu Zhang, Baotian Hu, Min Zhang
2025-04-10
Summary
This paper talks about CIGEval, a smart AI system that checks how well AI-generated images follow specific instructions, like matching text descriptions or editing parts of a photo.
What's the problem?
Current tools for judging AI-made images either work only for certain tasks or don’t explain why they give a score, making it hard to trust their results.
What's the solution?
CIGEval uses a mix of AI tools to break down image checks into smaller steps, like checking if edits match instructions or if details stay consistent, and combines these checks into a final score that aligns with human opinions.
Why it matters?
This helps make AI image tools more reliable for tasks like designing ads or editing photos, ensuring they follow instructions accurately and reducing the need for manual checks.
Abstract
Conditional image generation has gained significant attention for its ability to personalize content. However, the field faces challenges in developing task-agnostic, reliable, and explainable evaluation metrics. This paper introduces CIGEval, a unified agentic framework for comprehensive evaluation of conditional image generation tasks. CIGEval utilizes large multimodal models (LMMs) as its core, integrating a multi-functional toolbox and establishing a fine-grained evaluation framework. Additionally, we synthesize evaluation trajectories for fine-tuning, empowering smaller LMMs to autonomously select appropriate tools and conduct nuanced analyses based on tool outputs. Experiments across seven prominent conditional image generation tasks demonstrate that CIGEval (GPT-4o version) achieves a high correlation of 0.4625 with human assessments, closely matching the inter-annotator correlation of 0.47. Moreover, when implemented with 7B open-source LMMs using only 2.3K training trajectories, CIGEval surpasses the previous GPT-4o-based state-of-the-art method. Case studies on GPT-4o image generation highlight CIGEval's capability in identifying subtle issues related to subject consistency and adherence to control guidance, indicating its great potential for automating evaluation of image generation tasks with human-level reliability.