Rare Animal Fine-Grained Visual Classification (FGVC) Benchmark

# Rare Animal Fine-Grained Visual Classification (FGVC) Benchmark

## Overview

This benchmark contains four fine-grained visual classification datasets for rare animal species: **Bear**, **Turtle**, **Python**, and **Panther**. It was developed to support research in low-shot learning environments, particularly for applications in biodiversity conservation and wildlife protection.

The primary challenge of this benchmark lies in the high morphological similarity between species within each group and the scarcity of data, reflecting real-world conditions. Notably, **33 out of the 40 species** included are not present in the ImageNet-1K dataset, making it a rigorous test for model generalization.

This dataset was created as part of the undergraduate thesis "Enhancing Rare Animal Image Classification Performance Using Diffusion-Based Augmentation" at Ton Duc Thang University.

**Authors:** Ly Phi Hoc, Le Quang Duy

**Advisor:** Mr. Tran Minh Tuan

**Institution:** Faculty of Information Technology, Ton Duc Thang University, Ho Chi Minh City, Vietnam.

## Dataset Structure

The benchmark is organized into four main directories, one for each animal group. Each group contains 10 distinct species.

-   **Total Animal Groups:** 4 (Bear, Turtle, Python, Panther)

-   **Total Species:** 40

-   **Images per Species:** 200

-   **Total Images:** 8,000

-   **Data Split:** Each species has 100 images for training and 100 for testing, located in `train` and `test` subdirectories, respectively.

The directory structure is as follows:

```

<ANIMAL_GROUP>/

├── train/

│   ├── <SPECIES_NAME_1>/

│   │   ├── image_001.jpg

│   │   └── ...

│   └── <SPECIES_NAME_2>/

│       └── ...

└── test/

├── <SPECIES_NAME_1>/

│   └── ...

└── <SPECIES_NAME_2>/

└── ...

```

## Collection and Curation

The images were meticulously collected to evaluate data augmentation strategies in low-shot scenarios. The selection process was guided by two main criteria:

1.  **Conservation Status:** Prioritizing species with a concerning global conservation status.

2.  **Morphological Similarity:** Choosing species within each group that exhibit subtle visual differences to create a challenging FGVC task.

## License

This dataset is made available under the **Creative Commons Attribution 4.0 International (CC BY 4.0)** license. You are free to share and adapt the material for any purpose, even commercially, as long as you give appropriate credit.