---
pretty_name: "Sarawak Wildlife Image Classification Dataset"
task_categories:
  - image-classification
tags:
  - image
  - wildlife
  - sarawak
  - computer-vision
size_categories:
  - n<1K
---

[简体中文](README.zh-CN.md)

# Sarawak Wildlife Image Classification Dataset

## Overview

This dataset contains 865 JPG camera-trap wildlife images organised for multi-class image classification. It was prepared for the COS30019 Introduction to Artificial Intelligence Assignment 2, Scenario A: Wildlife Species Classification in Sarawak's National Parks using Camera Trap Images.

The dataset uses one folder per class. The folder name is the class label used by the project models.

## Dataset Summary

| Item | Value |
| --- | --- |
| Task | Multi-class image classification |
| Images | 865 unique JPG images |
| Classes | 10 |
| On-disk size | Approximately 362 MB |
| Image labels | Parent folder names |
| Current included split | Training only |

## Classes and Distribution

| Class label | Images |
| --- | ---: |
| bearded pig | 161 |
| leopard cat | 57 |
| long tailed macaque | 56 |
| malayan civet | 224 |
| marbled cat | 50 |
| munjact | 42 |
| pig tailed macaque | 55 |
| sambar deer | 91 |
| sun bear | 63 |
| yellow throated marten | 66 |
| **Total** | **865** |

`munjact` is retained exactly as the original project folder label. Changing it would change the label mapping used by the existing models.

## Directory Structure

```text
dataset/
├── README.md
├── README.zh-CN.md
└── train/
    ├── bearded pig/
    ├── leopard cat/
    ├── long tailed macaque/
    ├── malayan civet/
    ├── marbled cat/
    ├── munjact/
    ├── pig tailed macaque/
    ├── sambar deer/
    ├── sun bear/
    └── yellow throated marten/
```

## Assignment Context and Compliance

The assignment scenario asks for wildlife classification in Sarawak using camera-trap images. It specifies at least five classes with 150 training images per class, and a typical development workflow with training, validation, and testing data.

This dataset has ten classes, but only `bearded pig` (161 images) and `malayan civet` (224 images) meet the 150-images-per-class training target. The remaining eight classes have fewer images. Therefore, the dataset does not yet meet that per-class requirement.

## Split Notice

The original project also contains a separate `val` directory outside this dataset folder. Its 178 images are all byte-for-byte duplicates of images in `train`, so it is not included here as an independent validation split.

Before reporting model performance, create non-overlapping training, validation, and test splits from the 865 unique images. Images from the same camera-trap sequence should preferably stay in the same split to reduce leakage from near-duplicate frames.

## Intended Use

This dataset is intended for:

- COS30019 coursework and demonstrations.
- Educational experiments in image classification.
- Prototype research on automated wildlife-image categorisation.

It must not be used as the only evidence for conservation, wildlife-management, law-enforcement, or safety decisions.

## Limitations and Biases

- The classes are strongly imbalanced, from 42 `munjact` images to 224 `malayan civet` images.
- The dataset is small for deep learning and may encourage overfitting.
- The images may not represent all locations, seasons, lighting conditions, camera types, poses, or animal ages in Sarawak.
- The dataset has no independent validation or test split in its current published structure.
- Performance on this dataset may not generalise to other wildlife datasets or real-world deployments.

## Data Source, Rights, and License

The image source, copyright holder, and redistribution permission have not been recorded in this repository. Do not publish the images publicly or select a Hugging Face license until the data owner and redistribution rights have been confirmed.

## Citation

If you use this dataset, cite the original image source and the COS30019 Assignment 2 project. Add a formal citation here once the original data source is confirmed.