Reproduce “CPJ: Explainable Agricultural Pest Diagnosis via Caption-Prompt-Judge with LLM-Judged Refinement” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Wentao Zhang, Tao Fang, Lina Lu, Lifei Wang, Weihe Zhong

A training-free Caption-Prompt-Judge (CPJ) framework enhances explainable agricultural pest diagnosis and knowledge answering by leveraging iteratively refined image captions and an LLM-as-a-Judge for robust answer selection. The approach demonstrated improved crop classification accuracy of 63.38% and a knowledge QA score of 84.5 points on the CDDMBench dataset, representing gains of 22.7 percentage points and 19.5 points, respectively, over baseline methods.

Key insights

  • Iteratively refining multi-angle image captions with an LLM-as-a-Judge significantly enhances the semantic density and objectivity of visual information, improving downstream VQA performance.
  • A training-free, few-shot framework can achieve substantial accuracy improvements in agricultural pest diagnosis, overcoming limitations of costly supervised fine-tuning and improving adaptability to domain shifts.
  • The "LLM-as-a-Judge" paradigm effectively filters hallucinations and verifies factual completeness in generated answers, leading to more reliable and trustworthy diagnostic recommendations.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.