pxGPT: A schema-constrained multimodal LLM framework for plant phenotyping.

pxGPT achieves 0.572 to 0.712 agreement with human annotators in extracting 38-45 plant traits from Brassica images, matching human consistency levels.

• Why it matters: Accurate, scalable phenotyping is essential for plant research and breeding, but current methods are limited by predefined traits or manual effort, creating a need for flexible, automated solutions.
• What they did: The authors developed pxGPT, a Python framework that converts multi-view plant images into structured phenotype records using schema-guided information extraction evaluated on three Brassica datasets with nearly 79,000 annotations.
• The result: pxGPT reliably extracts visible plant traits without separate model training, with false-positive rates of 0.5-8.0%, and demonstrates consistent performance across different models, supporting its use for scalable plant phenotyping.