README.md

metadata

license: cc-by-4.0
task_categories:
  - image-text-retrieval
language:
  - en
tags:
  - agriculture
  - plant-disease
  - image-text-retrieval
  - leafnet
  - structured-captions
pretty_name: LeafNet 5-Caption Structured
size_categories:
  - 10K<n<100K

LeafNet 5-Caption Structured Dataset

Description

Agricultural leaf image dataset with 5-angle structured captions for image-text retrieval.

Each image has 5 captions from different perspectives:

• [Distribution]: Lesion/spots distribution and quantity
• [Lesion]: Individual lesion shape, border, and internal structure
• [Color]: Color changes, halos, powdery coatings, or texture
• [Location]: Where symptoms concentrate on the leaf
• [Diagnosis]: Comparison to healthy state and distinguishing features

Disease names are NOT included in captions to ensure genuine cross-modal understanding.

Stats

• Images: 13,012
• Classes: 89 (crop-disease pairs)
• Captions per image: 5
• Total captions: 65,060

Splits

Split	Images
Train	10,405
Val	1,299
Test	1,308

Files

• : All 13,012 images with 5 captions each
• : Training set (1 caption per image)
• : Validation set (1 caption per image)
• : Test set (1 caption per image)
• : Test set with all 5 captions per image
• : img2txt and txt2img mapping for retrieval evaluation
• : Split details and class distribution