meed2000/fungi-id

🍄 Fungi ID Dataset

This dataset contains 2,829 curated images of fungal species, specifically prepared to support the FungiTastic Identification System web application. It provides one representative image per species, optimized for browser-based visual comparison and reference.

🔍 Dataset Purpose

This dataset is not intended for model training. Instead, it is designed to be used alongside the FungiTastic-optimized Swin Transformer model for:

• Visual Verification: Displaying reference images next to user uploads to confirm AI predictions.
• Species Browsing: Allowing users to explore all 2,829 supported fungal species with high-quality examples.
• Providing Metadata: Supplying scientific names and stable IDs (category_id) that match the model's output indices.

🗂️ Dataset Structure

The dataset is organized for easy integration into web applications. It contains the following columns:

Column Name	Type	Description
image	Image	The representative JPEG image for the species (resized to 500px width).
category_id	int32	The numerical class ID used by the trained model (range 0-2828).
species	string	The common binomial species name (e.g., "Amanita muscaria").
scientificName	string	The full scientific name with authority (e.g., "Amanita muscaria (L.) Lam.").
filename	string	The original filename from the source FungiTastic dataset.

📦 Additional Files

Along with the main dataset, the repository includes essential JSON files for your web application:

• mushroom_labels.json: A direct mapping from category_id to scientificName. Use this to interpret the model's numerical predictions.

  {
    "0": "Agaricus augustus",
    "1": "Agaricus campestris",
    ...
    "2828": "Sarea resinae"
  }
  

• species_mapper.json: A comprehensive file containing all samples with their metadata and relative image paths, useful for populating browse interfaces and lookups.

🚀 Usage with the FungiTastic Web App

This dataset is designed to be used directly from Hugging Face by the FungiTastic web application.

1. Load the Dataset in the Browser (e.g., using the Hugging Face Hub API or by fetching the JSON mappers):

   // Example: Fetch the species mapper
   const mapperResponse = await fetch('https://huggingface.co/datasets/meed2000/fungi-id/raw/main/species_mapper.json');
   const speciesMapper = await mapperResponse.json();
   
   // Fetch the label map for model outputs
   const labelsResponse = await fetch('https://huggingface.co/datasets/meed2000/fungi-id/raw/main/mushroom_labels.json');
   const mushroomLabels = await labelsResponse.json();
   

2. Access Images: You can construct image URLs directly.

   https://huggingface.co/datasets/meed2000/fungi-id/resolve/main/images/0-3032624318.JPG
   

🛠️ Building the Dataset

This dataset was curated from the larger FungiTastic Kaggle dataset using a this process:

1. Download: Acquired the raw data from Kaggle.
2. Deduplication by Species: Processed the training metadata to select one representative image per category_id, ensuring coverage for all 2,829 classes in the model's closed set.
3. Mapping Generation: Created the JSON mapper files to link filenames, IDs, and scientific names.
4. Upload: The final set of images and metadata was uploaded to Hugging Face Hub for easy, public access.

The full preparation pipeline, managed with UV and a Justfile, is available in the associated fungi-id GitHub repository.

⚖️ License

This dataset is made available under the Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) license.

• You are free to share and adapt the material for any non-commercial purpose, as long as you give appropriate credit.
• You may not use the material for commercial purposes.

For the full license text, please visit: https://creativecommons.org/licenses/by-nc/4.0/

🙏 Acknowledgements

• The original FungiTastic dataset creators on Kaggle for compiling this comprehensive resource.
• BVRA for providing the fine-tuned Swin Transformer models that define the class set used here.

---

⭐ If you use this dataset, please consider starring the repository!