2026
A multiregional image-text dataset and benchmark for vision-language modeling of plant diseases
Trang V. Nguyen*#, Khang Nguyen Quoc*, David Harwath, Luyl-Da Quach, Phuong D. Dao# (* equal contribution, # corresponding author)
Under review. 2026
Plant disease datasets are often fragmented, dominated by controlled-environment imagery, and insufficient for developing transferable AI systems. Here, we present LeafMD, a multimodal resource combining LeafNet 2.0—255,855 image–text pairs spanning 37 crops, 197 crop–disease classes, and nine geographic regions—with LeafBench 2.0, a benchmark evaluating fine-grained pathological reasoning across nine tasks. Evaluation of 16 vision–language models reveals substantial gaps between disease recognition and symptom-level reasoning, highlighting LeafMD’s value for developing disease-aware agricultural foundation models.
A multiregional image-text dataset and benchmark for vision-language modeling of plant diseases
Trang V. Nguyen*#, Khang Nguyen Quoc*, David Harwath, Luyl-Da Quach, Phuong D. Dao# (* equal contribution, # corresponding author)
Under review. 2026
Plant disease datasets are often fragmented, dominated by controlled-environment imagery, and insufficient for developing transferable AI systems. Here, we present LeafMD, a multimodal resource combining LeafNet 2.0—255,855 image–text pairs spanning 37 crops, 197 crop–disease classes, and nine geographic regions—with LeafBench 2.0, a benchmark evaluating fine-grained pathological reasoning across nine tasks. Evaluation of 16 vision–language models reveals substantial gaps between disease recognition and symptom-level reasoning, highlighting LeafMD’s value for developing disease-aware agricultural foundation models.