Beans

Overview

The Beans dataset (also known as IBeans) contains leaf images for classification of bean plant diseases. The dataset represents three classes: healthy leaves, angular leaf spot disease, and bean rust disease. Images were collected in Uganda for practical disease classification in the field.

• Train: 1,034 images

• Validation: 133 images

• Test: 128 images

This dataset is particularly valuable for developing agricultural AI applications to help farmers identify and manage crop diseases early.

Data Structure

When accessing an example using ds[i], you will receive a dictionary with the following keys:

Key	Type	Description
image	PIL.Image.Image	RGB image of bean leaf (variable dimensions, typically ~500x500)
label	int	Disease class label (0-2)

Disease Classes

The dataset includes 3 disease classes:

Label	Class Name
0	healthy - Healthy bean leaves with no disease
1	angular_leaf_spot - Angular leaf spot disease caused by Pseudocercospora griseola
2	bean_rust - Bean rust disease caused by Uromyces appendiculatus

Usage Example

Basic Usage

from stable_datasets.images.beans import Beans

# Load different splits
train_ds = Beans(split="train")
val_ds = Beans(split="validation")
test_ds = Beans(split="test")

# Inspect a sample
sample = train_ds[0]
print(sample.keys())  # {"image", "label"}
print(f"Label: {sample['label']}")  # 0, 1, or 2

# Optional: make it PyTorch-friendly
train_torch = train_ds.with_format("torch")

Getting Class Names

from stable_datasets.images.beans import Beans

ds = Beans(split="train")

# Get class names
class_names = ds.info.features["label"].names
print(class_names)  # ["healthy", "angular_leaf_spot", "bean_rust"]

# Map label to class name
sample = ds[0]
disease_name = class_names[sample["label"]]
print(f"Disease: {disease_name}")

Typical Workflow: Disease Classification

from stable_datasets.images.beans import Beans
from torchvision import transforms
from torch.utils.data import DataLoader

# Load and prepare data
train_ds = Beans(split="train")
val_ds = Beans(split="validation")

# Convert to PyTorch format
train_ds = train_ds.with_format("torch")
val_ds = val_ds.with_format("torch")

# Apply transforms
transform = transforms.Compose([
    transforms.Resize((224, 224)),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.485, 0.456, 0.406],
                       std=[0.229, 0.224, 0.225])
])

train_ds.set_transform(lambda x: {
    "image": transform(x["image"]),
    "label": x["label"]
})

# Create dataloaders
train_loader = DataLoader(train_ds, batch_size=32, shuffle=True)
val_loader = DataLoader(val_ds, batch_size=32)

# ... train your disease classification model ...

Dataset Statistics

The dataset is relatively small but balanced across the three classes, making it suitable for:

• Transfer learning and fine-tuning pre-trained models

• Benchmarking few-shot learning algorithms

• Educational purposes in agricultural AI

• Prototyping crop disease detection systems

References

• Homepage: https://github.com/AI-Lab-Makerere/ibean/

• License: MIT License

• Source: Makerere AI Lab, Uganda

Citation

@misc{makerere2020beans,
  author = "{Makerere AI Lab}",
  title = "{Bean Disease Dataset}",
  year = "2020",
  month = "January",
  url = "https://github.com/AI-Lab-Makerere/ibean/"
}