deepplants/cabbage

🥬 CABBAGE: Comprehensive Agricultural Benchmark Backed by AI-Guided Evaluation

**🌐 Homepage** | **🏆 Leaderboard** | **🤗 Dataset** | **💻 GitHub**

Table of Contents

• —CABBAGE: Comprehensive Agricultural Benchmark Backed by AI-Guided Evaluation
• —Table of Contents
• —Dataset Description
• —Dataset Summary
• —Supported Tasks and Leaderboards
• —🖼️ Agronomic Visual Cognition
• —📚 Agricultural Scientific Knowledge
• —🛠️ Agricultural Procedural Reasoning
• —Loading the Dataset

Dataset Description

• —Homepage: https://huggingface.co/datasets/boilnserve/cabbage
• —Repository: https://github.com/boilnserve/cabbage
• —Paper: Not yet published
• —Leaderboard: https://huggingface.co/datasets/boilnserve/cabbage
• —Size of downloaded dataset files: 3.74 GB
• —Size of the auto-converted Parquet files: 240.84 MB
• —Number of rows: 74,206

CABBAGE is a large-scale, multimodal benchmark for evaluating AI systems in agriculture across three complementary task macro-categories: Visual Cognition, Scientific Knowledge, and Procedural Reasoning. Each macro-category contains high-quality, domain-specific subsets built from curated or expert-reviewed sources.

Benchmark Tracks and Subsets

<img src="https://cdn-uploads.huggingface.co/production/uploads/6824cf491019386a26b831c1/PYEaWom37vSKPDfKqt7fo.png" width="600"/>

🖼️ Agronomic Visual Cognition

Evaluates image-based plant understanding and visual QA.

• —`eppo`: Plant pest and disease image classification (from EPPO data)
• —`plantnet`: Species-level classification using Pl@ntNet imagery
• —`bppq`: The Big Plant Pathology Quiz — visual QA for pathology

📚 Agricultural Scientific Knowledge

Tests scientific factual knowledge, retrieval, and reasoning over structured agronomic data.

• —`agriexam`: Multiple-choice exams from official agricultural education materials
• —`cca_ceu`: Multiple-choice questions from the Certified Crop Adviser datasets and Continuing Education Unit materials
• —`embrapa`: Questions derived from Brazilian Agricultural Research Corporation technical guide series

🛠️ Agricultural Procedural Reasoning

Challenges models on procedural tasks derived from domain-relevant manuals and wikiHow entries.

• —`wikihow_arrange`: Arrange steps of an agricultural procedure in the correct order
• —`wikihow_missing`: Identify missing steps in an agricultural workflow
• —`wikihow_next`: Predict the next step in a given task
• —`wikihow_all`: Generate the full sequence of steps required to carry out an agricultural task

Loading the Dataset

You can load any specific configuration and split using the Hugging Face datasets library:

from datasets import load_dataset

# Example: Load all the splits from the Agronomic Visual Cognition subset
dataset_dict = load_dataset("deepplants/cabbage", name="agronomic_visual_cognition")

# Example: Load the Embrapa split from the Agricultural Scientific Knowledge subset
dataset = load_dataset("deepplants/cabbage", name="agricultural_scientific_knowledge", split="embrapa")