AlYldz/Gradio

Sheep Identification Dataset (172 Classes)

This dataset is specifically curated for individual sheep identification tasks. It contains processed images of 172 different sheep IDs, derived from a Mendeley Data source and optimized for deep learning models.

1. Project Overview

Individual identification is a critical task in precision livestock farming. This dataset provides a multi-class classification structure where each class represents a unique sheep ID (labeled from koyun1 to koyun172).

2. Dataset Statistics

The dataset is structured to support robust training and objective evaluation:

• —Number of Classes: 172 unique sheep identities.
• —Total Images: 14,810 images.
• —Training Set: 14,443 samples (including augmented data).
• —Validation Set: 367 samples.

3. Preprocessing & Engineering (The "Pipeline")

To handle the complexity of 172 different classes, the following preprocessing steps were implemented:

• —Image Standardization: All images were resized and normalized to ensure consistent input for CNN/Transformer architectures.
• —Data Augmentation: Given the high number of classes, we applied augmentation to the training set to prevent overfitting and improve the model's ability to recognize sheep from different angles:
• —Rotation & Flips: To simulate different camera perspectives in the field.
• —Brightness/Contrast: To account for varying lighting conditions in farm environments.
• —Data Splitting: A strict separation between training and validation sets was maintained to ensure the integrity of the performance metrics.

4. Source & Citation

• —Original Data Source: https://data.mendeley.com/datasets/cstwjgtxfd/
• —Original Authors: Sanabel Abu Jwade,Andrew Guzzomi,Ajmal Mian,Galib Muhammad Shahriar Himel,Md Masudul Islam
• —Modifications: Data re-structured into ImageFolder format, augmented, and split into train/val by @azaliyasli.

5. How to Load

from datasets import load_dataset
dataset = load_dataset("AlYldz/Gradio")