Apppublic

Nausad/Vision_Transformer_Based_Image_Classification_Model

sourceHugging Facemitupdated 4mo agoView on Hugging Face

0likes

App README

🔍 CIFAR-10 Vision Transformer Classifier

A web-based image classification application powered by a custom Vision Transformer (ViT) trained from scratch on the CIFAR-10 dataset. Upload any image and get real-time predictions across 10 categories.

🚀 Demo

Upload any image via drag-and-drop or file browser. The model resizes it to 32×32 pixels (the native CIFAR-10 resolution) and predicts one of 10 classes with confidence scores.

Supported Classes: airplane · automobile · bird · cat · deer · dog · frog · horse · ship · truck

🧠 Model Architecture

Component	Details
Architecture	Vision Transformer (ViT) — built from scratch
Patch Size	4 × 4 pixels
Embedding Dim	256
Transformer Depth	20 layers
Attention Heads	8
MLP Dim	512
Total Parameters	~10.57 Million
Input Size	32 × 32 × 3
Output Classes	10

Training Details

Metric	Value
Dataset	CIFAR-10 (50K train / 10K test)
Epochs	25
Training Accuracy	72.59%
Test Accuracy	70.51%

🛠️ Tech Stack

—Backend: Flask (Python)
—ML Framework: PyTorch
—Frontend: HTML5 · Vanilla CSS · JavaScript
—Model: Custom Vision Transformer (no pretrained weights)
—Containerization: Docker

📁 Project Structure

├── flask_app.py          # Flask server (routes: /, /predict, /model-info)
├── cifar10_model.pth     # Trained model weights
├── requirements.txt      # Python dependencies
├── Dockerfile            # Container configuration
└── templates/
    └── index.html        # Full-featured web UI

⚙️ Run Locally

bash

# 1. Install dependencies
pip install -r requirements.txt

# 2. Start the Flask server
python flask_app.py

# 3. Open in browser
# → http://127.0.0.1:5000

🐳 Run with Docker

bash

# Build the image
docker build -t cifar10-vit .

# Run the container
docker run -p 7860:7860 cifar10-vit

# Open in browser
# → http://localhost:7860

📊 Features

—🖼️ Drag & Drop upload — supports JPEG, PNG, WEBP, BMP
—🔲 32×32 pixel simulator — see exactly what the model processes
—📊 Animated probability bars for all 10 classes
—⚡ Real-time inference with latency display (milliseconds)
—🎨 Warm beige premium UI with glassmorphism cards

👤 Author

Nausad Ali Built with ❤️ and a Cup of Coffee

📄 License

This project is licensed under the MIT License.