Infraspecific Plant Vision Benchmarking Dataset: Banana, Grapevine, and Market Garden Crops

LICENSE NOTICE: This repository contains images under various licenses (CC-BY-SA, CC-BY, and All Rights Reserved / Research Use Only). The Zenodo metadata license does NOT override individual image rights. Users MUST refer to the license column in the provided train_val.csv and test.csv files to determine the exact allowed use for each specific image.

# Infraspecific Plant Vision Benchmarking Dataset

## Description
This Zenodo repository provides three image datasets (Banana, Grapevine, Market Garden) designed for the evaluation of computer vision models on the infraspecific identification (varieties and cultivars) of plants.

## Archive Organization
The repository contains three uncompressed `.tar` archives, corresponding to each evaluated crop:

1. `banana_dataset.tar`: Varieties and wild species of the Ensete, Musa and Musella genera.
2. `grape_variety_dataset.tar`: Grapevine varieties (Vitis vinifera L.).
3. `plant_market_garden_dataset.tar`: Market garden crop varieties (aubergines, courgettes, tomatoes).

Each extracted archive follows an identical internal structure:

* `train/` directory containing subdirectories for each class with their respective training images.
* `test/` directory containing flat test images (for blind evaluation protocols).
* `train_val.csv`: Metadata for the training set.
* `test.csv`: Metadata for the test set.
* `class_mapping.txt`: List of classes in the dataset.

## File Nomenclature
All images have been standardized (`.jpg` format) and renamed according to the following convention:
`[dataset_name]_[class_name]_[global_index].jpg`

Example: `market_cucurbita_pepo_ronde_de_nice_00150.jpg`

## Metadata Structure (CSV Files)
The `train_val.csv` and `test.csv` files provide the annotations and traceability for each image. They share the following column structure:

* **image_name**: Standardized image file name.
* **observation_id**: Unique identifier of the physical observation (iNaturalist number for bananas, incremental identifier for grapevine and market garden crops).
* **class_name**: Standardized name of the class (variety or taxon).
* **organ**: Photographed plant organ (e.g., leaf, fruit, habit, flower).
* **split**: Dataset assignment (`train` or `test`).
* **author**: Credited author or institution for the photograph.
* **license**: Distribution license applicable to the image.
* **source_url**: Link to the source data or institutional portal.
* **date**: Date the photograph was taken in ISO 8601 format (YYYY-MM-DD).
* **latitude / longitude**: GPS coordinates (if available).

## Sources and Licenses
Due to the compilation of data from various sources, usage rights vary depending on the datasets and individual images. The `license` column in the CSV file is authoritative for each image.

### 1. Banana Dataset
* **Crowdsourced data (iNaturalist)**: Variable Creative Commons licenses (CC-BY, CC-BY-NC, etc.) depending on the original authors' choices.
* **Expert data (CIRAD)**: Property of the expert collection, use restricted to scientific research (All Rights Reserved - Research Use Only).

### 2. Grape Variety Dataset
* **Train split**: Images produced as part of experimental campaigns by Pl@ntNet consortium (Inria, CIRAD, IRD, INRAE, CNRS, Univ. Montpellier). Distributed under the CC-BY-SA license.
* **Test split (PlantGrape)**: Images from the institutional portal of IFV, INRAE, and L'Institut Agro. Use restricted to scientific research (All Rights Reserved - Research Use Only).

### 3. Plant Market Garden Dataset
* **Entire dataset**: Images produced as part of experimental campaigns by Pl@ntNet consortium (Inria, CIRAD, IRD, INRAE, CNRS, Univ. Montpellier). Distributed under the CC-BY-SA license.