Context-Aware Multimodal Augmented PlantVillage Dataset

We present a multimodal dataset designed for crop disease detection that pairs high-resolution plant disease images with rich textual symptom descriptions and contextual metadata. The dataset extends the widely used PlantVillage collection by incorporating over 3,900 expert-curated text prompts, covering 38 disease classes across 14 crop species. Each image is annotated with detailed narratives describing lesion morphology, symptom progression, and environmental factors, along with structured fields including pathogen type, soil conditions, and climatic ranges. This dataset aims to advance research in vision-language models for agriculture by enabling joint learning of visual and textual cues, improving generalization to real-world farm conditions, and facilitating robust, explainable plant disease classification.