Wildlife Image Classifier

---

📸 Project Overview

This is an AI Powered Image Classification System that identifies wild Aniamls like:

• Cheetah
• Leopard
• Tiger

Using Deep Learning + Computer Vision + Transfer Learning, this project leverages EfficientNetV2-S — one of the most state-of-the-art CNN architectures — to deliver high-accuracy animal identification.

---

🎯 Key Features

Feature	Description
State-of-the-art Model	Uses EfficientNetV2-S architecture for high accuracy and efficient training
Transfer Learning	Pre-trained on ImageNet, fine-tuned on custom wildlife dataset
Data Augmentation	Increases dataset diversity with random flips, rotations, and color jitter
Early Stopping	Prevents overfitting during training
TorchScript Export	Model saved in TorchScript format for fast and optimized inference
Fully Automated Inference Pipeline	Automatically handles preprocessing, model loading, and prediction
Web Deployment (Flask)	Flask backend handles image uploads and predictions
Web UI	Clean and modern interface to upload images and get predictions
End-to-End Deployment	Can run locally or be deployed to cloud platforms

---

🔧 Tech Stack

Module	Description
PyTorch	Deep Learning Framework
EfficientNetV2-S	Transfer Learning Backbone
Albumentations	Advanced Data Augmentations
TorchScript	Model Export Format for Inference
Flask	Backend Web Server
HTML/CSS	Frontend Web Interface
OpenCV	Image Preprocessing

---

📊 Model Pipeline

1️⃣ Input Image (Cheetah, Leopard, Tiger):

• User uploads an image of any of the 3 target animals.
• This is the raw data entry point.

2️⃣ Preprocessing Pipeline:

• Resize image to 224x224 pixels — required input size for EfficientNetV2-S.
• Normalize using ImageNet statistics (mean/std).
• Ensures consistency with pretrained model expectations.

3️⃣ EfficientNetV2-S Model (Core Deep Learning Engine):

Uses:

• CNN backbone (convolutional layers extract visual features: textures, patterns, shapes).
• Global Average Pooling (reduces dimensionality while preserving spatial features).
• Fully Connected Layer (classification layer with 3 outputs for 3 classes).
• Softmax Activation (produces class probabilities).

4️⃣ Output Predictions:

Model returns:

• Class label (Cheetah, Leopard, Tiger).
• Confidence score (ex: 93% confidence it’s a Tiger).

5️⃣ Flask Deployment API:

Backend server:

• Receives uploaded images.
• Applies preprocessing.
• Runs inference using the TorchScript exported model.
• Sends prediction results back to UI.

6️⃣ Frontend UI:

• User interface where:
• Users upload images.
• View predictions.
• Clean, futuristic design for smooth interaction.

---

⚙️ Setup Instructions

1️⃣ Clone the repository:

git clone https://github.com/SrujanPR/Animal-Classification.git
cd Animal-Classification

2️⃣ Install Dependencies

pip install -r requirements.txt

3️⃣ Model Training

There is EfficientNetV2_Model.ipynb python notebook in this project. You can copy use that to train the model from scratch using your own dataset in google colab and save the trained model file.

I have already completed the training with a dataset of 5000 images and I have also saved the best trained model file as efficientnetv2_model2_inference.pt. This is also available in this project directory.

4️⃣ Setup the FastAPI backend server:

run the flask app using the command:

python app.py

5️⃣ Access the Web App

Open your browser and go to this URL: (http://127.0.0.1:5000) The webapp will open and you can use this.

---

📸 Demo Screenshot

---

💡 Future Improvements

• 🔬 Add more animal classes 🐻 🐘 🦁
• 📱 Build a mobile version (Flutter + TensorFlow Lite)
• 🧮 Add Continuous Model Training Pipeline (MLOps)

---

📜 License

This project is licensed under the MIT License.

---

📢 Credits

• Built by SRUJAN PR 🧠
• Model powered by EfficientNetV2-S
• Inspired by the beauty of wildlife 🌿
• Feel free to reach out for collaborations, ideas, or improvements. LinkedIn

---

⭐ Star this repo if it helped you!

---