anhaltai/plantNaturalist500k

Dataset Card for plantNaturalist-500k

<center><img src="blessed-sunburst.png"></img></center>

Dataset Description

<!--

• —Repository: (TODO)
• —Paper: (TODO)
• —Point of Contact: Serhii Hamotskyi, Christian Hänig -->

Dataset Summary

plantNaturalist-500k is a large-scale dataset for in-the-wild plant species classification, real-world environments.

It contains 500,909 images spanning 2,491 plant species, sampled from iNaturalist Research-grade Observations.

The dataset is specifically designed to serve as a benchmark for evaluating plant classification models in real-world scenarios, complementing existing datasets like Pl@ntNet-300k. It features a diverse range of plant species captured in various environmental conditions, angles, and lighting situations, making it particularly suitable for training and evaluating robust plant classification models.

Supported Tasks and Leaderboards

image-classification: The dataset can be used to train and evaluate a model for plant species classification, which consists in identifying the correct plant species from an image.

Dataset Structure

Data Instances

An instance in the dataset consists of:

• —An image of a plant in its natural environment
• —Taxonomic information (kingdom, phylum, class, order, family, genus, species)
• —Geographic location (latitude and longitude)

For example:

{
  "image": "path/to/image.jpg",
  "gbifID": "3355189127",
  "species": "Rosa canina",
  "genus": "Rosa",
  "family": "Rosaceae",
  "order": "Rosales",
  "class": "Magnoliopsida",
  "phylum": "Tracheophyta",
  "kingdom": "Plantae",
  "decimalLatitude": 41.41389465332031,
  "decimalLongitude": -74.32788848876953,
}

Data Splits

The dataset is split into three subsets:

• —Training set: 300,545 images
• —Validation set: 100,182 images
• —Test set: 100,182 images

The splits are created by randomly dividing the observations (each containing one or more photos) in proportions of 60%/20%/20%.

Dataset Creation

Curation Rationale

The dataset was created to address the need for a high-quality benchmark dataset for plant species classification in real-world environments. It was specifically designed to:

1. 1.Provide a large, diverse set of plant images from various geographical locations
2. 2.Ensure balanced representation of species across different regions
3. 3.Enable reproducible evaluation through predefined train/validation/test splits

Source Data

The dataset was compiled from iNaturalist Research-grade Observations, using a GBIF export from December 23, 2023: <https://www.gbif.org/occurrence/download/0034217-231120084113126>.

Additional Information

Licensing Information

The dataset is released under the Deed - Attribution-NonCommercial 4.0 International - Creative Commons license.

<!--

Citation Information

TODO -->