Open Evaluation Agent: Efficient and Promptable Evaluation of Visual Generative Models

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Shulin Tian, Ziqi Huang, Fan Zhang, Hongyuan Zhu, Yu Qiao, Ziwei Liu
• Venue: arXiv: cs.AI
• Published: 2026-08-10
• Collected: 2026-08-11
• URL: http://arxiv.org/abs/2608.09666v1
• PDF: https://arxiv.org/pdf/2608.09666v1
• DOI: UNKNOWN
• Registry group: academic_apis_and_preprints
• Evidence role: canonical_academic_index
• Language: en
• Country / region: UNKNOWN

Plain-Language Brief

Researchers propose a new framework called Evaluation Agent to quickly and efficiently evaluate visual generative models. The agent uses human-like strategies for multi-round evaluations, providing detailed, user-tailored analyses based on natural language requests. Experiments show that the agent reduces evaluation time by 90% compared to traditional methods while delivering comparable results.

What the Document Says

• The Evaluation Agent is a framework designed for efficient and dynamic multi-round evaluations of visual generative models.
• It uses human-like strategies to decompose natural-language evaluation requests, generate targeted prompts, sample images or videos from the evaluated model, invoke suitable evaluation tools, and iteratively update its plan based on observed evidence.
• The framework covers both predefined benchmark dimensions and open-ended user concerns.
• Experiments show that the Evaluation Agent reduces evaluation time to 10% of traditional methods while delivering comparable results.
• Open Evaluation Agent (Open-EA) is introduced, which is based on a local planning backbone trained from Qwen2.5-3B-Instruct.

Key Concepts Mentioned

• Evaluation Agent
• Visual Generative Models
• Natural Language Requests
• Multi-round Evaluations
• Evaluation Tools
• Open Evaluation Agent (Open-EA)

Claims Made by the Document

• The Evaluation Agent is efficient, promptable, explainable, and scalable.
• Open-EA shows partial cross-family transfer of the learned policy.

Evidence Used

• The text describes the features of the Evaluation Agent.
• Experiments are mentioned to validate the claims about the agent.

Methods / Approach

• Decomposition of natural language requests
• Generation of targeted prompts
• Sampling images or videos from evaluated models
• Invocation of evaluation tools
• Iterative updating of plans based on observed evidence

Limitations Stated by the Authors

Not stated in source

Limitations Noted by the Agent

Agent interpretation — not from the source.

The agent's performance might depend on the quality and variety of the data used for training.

Open Questions

• What specific human-like strategies does the Evaluation Agent employ?

Notes for Future Q&A

May include interpretation. Labeled where it does.

The text mentions that humans quickly form impressions of a model's capabilities from only a few samples, but it does not explain in detail what human-like strategies the Evaluation Agent employs.

Raw Abstract / Excerpt

Recent advances in visual generative models have enabled high-quality image and video generation, but evaluating these models often demands sampling hundreds or thousands of images or videos, which is computationally expensive. Existing evaluation methods also rely on rigid pipelines that overlook specific user needs and provide numerical results without clear explanations. Mimicking how humans quickly form impressions of a model's capabilities from only a few samples, we propose the Evaluation Agent framework, which employs human-like strategies for efficient, dynamic, multi-round evaluations, offering detailed, user-tailored analyses. Given a natural-language evaluation request, the agent decomposes it into sub-aspects, generates targeted prompts, samples images or videos from the evaluated model, invokes suitable evaluation tools, and iteratively updates its plan from the observed evidence, covering both predefined benchmark dimensions and open-ended user concerns. The framework is thus efficient, promptable, explainable, and scalable across models and tools. Experiments show that Evaluation Agent reduces evaluation time to 10% of traditional methods while delivering comparable results. We further introduce Open Evaluation Agent (Open-EA) by constructing EA-CoT-10K, a corpus of history-conditioned step-level instruction-tuning records derived from multi-round evaluation rollouts, and training EA-3B from Qwen2.5-3B-Instruct as a local planning backbone that preserves the structured reasoning, tool invocation, and summary protocol of the API-based agent while reducing dependence on proprietary backbones. Experiments validate the API-based agent on established T2I/T2V benchmarks and open-ended queries, and evaluate Open-EA on four in-domain and three out-of-domain T2V generator families, showing partial cross-family transfer of the learned policy.