avrashmi/Animal_Detection_using_Machine_learning

Animal detection and classification of African wildlife (Buffalo, Elephant, Rhino, Zebra) using deep learning with PyTorch Lightning and Hugging Face’s YOLOS-tiny model. # Animal_Detection_using_Machine_learning Animal detection and classification of African wildlife (Buffalo, Elephant, Rhino, Zebra) using deep learning with PyTorch Lightning and Hugging Face’s YOLOS-tiny model. **Automated Animal Detection Using Machine Learning** This project implements an **object detection pipeline** to automatically detect and classify African wildlife (Buffalo, Elephant, Rhino, and Zebra) using **deep learning**. It leverages **transformers, PyTorch Lightning, and Hugging Face’s YOLOS-tiny model** for efficient detection and training on custom datasets. **Project Overview** The goal of this project is to **automatically detect and classify wild animals from images** using **state-of-the-art object detection models**. The workflow includes: 1. **Dataset Preparation** – Images and YOLO-style bounding box annotations of animals (Zebra, Buffalo, Rhino, Elephant). 2. **Data Processing** – Custom PyTorch Dataset class for reading, resizing, and normalizing images. 3. **Model Training** – Fine-tuning Hugging Face’s **YOLOS-tiny** object detection transformer using **PyTorch Lightning**. 4. **Evaluation** – Using **Mean Average Precision (mAP)** metrics at multiple IoU thresholds. 5. **Visualization** – Bounding box predictions drawn on sample images for visual inspection. **Tools & Technologies Used** - **Python 3.x** → Core programming language. - **PyTorch** → Deep learning framework for building datasets, models, and training. - **Torchvision** → Image transformations and utilities for working with vision datasets. - **PyTorch Lightning** → High-level framework for structuring and training deep learning models. - **Transformers (Hugging Face)** → Pretrained models like **YOLOS-tiny** for object detection. - **torchmetrics (detection)** → For computing **Mean Average Precision (mAP)** and evaluation metrics. - **timm** → Image models collection used as backbones for transformers. - **Datasets (Hugging Face)** → For data handling and prepro …