Image Prediction API

This repository provides a Flask-based API to perform image classification using a pre-trained model.

The model was trained using code from a separate repository:
🔗 Image-Classification

---

Model Training Overview

The models served by this API were trained using a deep learning pipeline for Intel Image Classification, developed in both TensorFlow and PyTorch.

Intel Image Classification Dataset

This dataset used to train the models is provided by Intel and is designed for natural scene classification. It contains RGB images across 6 natural categories:

• Buildings

• Forest

• Glacier

• Mountain

• Sea

• Street

Training Highlights

• TensorFlow Model
  • Data Augmentation: RandomFlip, RandomZoom, RandomRotation
  • CNN with BatchNorm, Dropout, and GlobalAveragePooling2D
  • Regularization via L1L2, EarlyStopping, ModelCheckpoint
• PyTorch Model
  • CNN with BatchNorm, Dropout, AdaptiveAvgPool2d
  • Best model checkpointed with torch.save

Evaluation Results

Framework	Train Acc	Val Acc	Train Loss	Val Loss
TensorFlow	84.25%	83.0%	0.5474	0.5870
PyTorch	87.01%	87.1%	0.3631	0.3473

Evaluation included:

• Classification Reports
• Confusion Matrices using Seaborn + Matplotlib

---

Project Structure

.
├── app.py                  # Flask app for serving image prediction API
├── main.py                 # Main script (could be CLI or init script)
├── request.py              # Internal/external request handler
├── model/
│   ├── bestmodel.keras     # TensorFlow model
│   └── bestmodel.pth       # PyTorch model
├── src/
│   ├── predict.py          # Prediction logic
│   └── __init__.py         # Python module initializer

---

Getting Started

1. Clone the Repository

git clone https://github.com/JAGAN-JGUN/image-prediction-api.git
cd image-prediction-api

2. Set Up Environment

python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate

3. Install Requirements

Install dependencies manually (if requirements.txt is missing):

pip install flask torch tensorflow pillow

4. Run the Flask App

python app.py

Visit: http://localhost:5000

---

Model Information

Both TensorFlow and PyTorch models are available:

• model/bestmodel.keras
• model/bestmodel.pth

Prediction logic is managed in src/predict.py. Switch between models based on your use case.

---

Example Prediction via curl

curl -X POST -F image=@sample.jpg http://localhost:5000/predict

---

Example prediction via CLI Script (request.py)

You can also make predictions directly from the command line using the request.py script. This is useful for testing locally without using the web UI.

Usage

python request.py path/to/image.jpg #Replace path/to/image.jpg with your image path.

This sends a POST request to the Flask API running at http://localhost:5000/predict.

It displays the selected image and prints the prediction result as JSON.

---

License

This project is licensed under the MIT License.

---

Acknowledgments

• Training code adapted from: Image-Classification
• Built with Flask, PyTorch, and TensorFlow.