deep-plants/deBary

Dataset Card for Dataset Name

Overview

Repository for the deBary plant disease dataset with train/test splits (85%/15%) proposed in [TBD].

Summary

• Images: 9,454 JPG files
• Train images: 6,065
• Test images: 3,389
• Species folders: 81
• Health statuses: healthy, diseased
• Condition labels: 1,265
• Climate CSV rows: 70,999

Image Layout

deBary/
├── data/
│   ├── train/
│   │   └── <species>/<health_status>/<condition>/*.jpg
│   └── test/
│       └── <species>/<health_status>/<condition>/*.jpg
├── climate_data.csv
├── filename_mapping.json
├── normalized_labels.json
└── split_manifest.json

Images are stored by split, species, health status, and condition:

data/<split>/<species>/<health_status>/<condition>/<file_name>.jpg

For healthy samples, the path omits the condition directory:

data/<split>/<species>/healthy/<file_name>.jpg

Examples:

data/train/chili_pepper/diseased/phytophagous/000001.jpg
data/train/chili_pepper/healthy/000001.jpg

Metadata Files

split_manifest.json

Lists each image assigned to a split. Each entry contains:

• file_name: normalized image filename, such as 000001.jpg
• rel_dir: path below the split directory, such as alfalfa/diseased/phytophagous
• split: train or test

filename_mapping.json

Maps normalized filenames back to their original source paths. Each entry contains:

• new_name: normalized filename used in this dataset
• rel_dir: normalized destination directory below the split
• source_rel_path: original relative source path
• split: train or test

normalized_labels.json

Contains mappings between original label strings and normalized slug labels used in the folder names.

Climate Data

climate_data.csv contains image-level visual features, geolocation, timestamps, and climate variables. The image column stores relative paths, for example:

../deBary/data/train/peach/healthy/000128.jpg

Main column groups:

• Labels: species, health_status, stress_condition, stress_id
• Image path: image
• Color features: mean_r, mean_g, mean_b, std_r, std_g, std_b, brightness_mean, brightness_std
• Location and time: latitude, longitude, altitude, timestamp, timestamp_local, month, hour, season, time_of_day, hour_slot
• Camera metadata: camera_model
• Weather grid: bbox
• Climate variables: max_temp, mean_temp, dewpoint_temp, humidity, precipitation

Usage Notes

• Join image files with climate_data.csv through the image column when climate or visual feature metadata is needed.
• The numeric label fields in climate_data.csv are encoded values; use the directory names and JSON metadata files when human-readable species or condition labels are required.