🔍 Neural Lens: High-Performance Image Recognition Engine

A full-stack Computer Vision web app built with Python (FastAPI) and Next.js. This tool utilizes MobileNetV2 to transform raw visual data into structured mathematical insights with high probabilistic precision, in a lightweight footprint suited for free-tier deployment.

---

🚀 Live Demo

Frontend: neurallens.vercel.app Backend API: neurallens-api.onrender.com

---

📺 Demo Preview

---

✨ Features

• Deep Residual Learning: Leverages a 50-layer ResNet architecture to overcome the vanishing gradient problem, ensuring high-fidelity feature extraction.
• Probabilistic Classification: Identifies over 1,000 object categories from the ImageNet dataset with a Top-5 confidence ranking system.
• Viewfinder UI: A custom "scanning" interface with animated corner brackets and a live scan-line effect while the model analyzes the image.
• Real-Time Inference: Optimized image preprocessing pipeline (224x224 RGB normalization) for fast classification via a dedicated API.
• Confidence Readout: Color-coded confidence tiers (high / medium / low) for both the top prediction and the full Top-5 breakdown.

---

🛠️ Tech Stack

• Backend: FastAPI (Python)
• Frontend: Next.js / React, Tailwind CSS
• Deep Learning Engine: TensorFlow (CPU build) / Keras
• Architecture: MobileNetV2 (Pre-trained on ImageNet)
• Deployment: Render (backend) + Vercel (frontend)

---

🚀 Installation & Local Setup

Backend

cd backend
pip install -r requirements.txt
uvicorn app.main:app --reload

Frontend

cd frontend
npm install
npm run dev

Create a .env.local file in frontend/ with:

NEXT_PUBLIC_API_URL=http://127.0.0.1:8000

---

📂 Project Structure

neurallens/
├── backend/
│   ├── app/
│   │   ├── main.py       # FastAPI routes
│   │   └── predict.py    # ResNet50 model loading and inference engine
│   └── requirements.txt
├── frontend/
│   └── src/app/
│       ├── page.js       # Upload UI, viewfinder animation, results
│       └── globals.css   # Scan-line and pulse animations
├── assets/
│   └── NeuralLens.gif
└── README.md

---

🧠 Model Insights

The engine utilizes MobileNetV2, an architecture specifically designed for efficient, low-memory inference — well suited for resource-constrained deployment environments.

• The Architecture: MobileNetV2 uses depthwise separable convolutions and inverted residual blocks to dramatically reduce parameter count and memory footprint compared to larger networks like ResNet50, while retaining strong classification accuracy.

• Input Transformation: Images are mathematically resized to 224x224x3 and normalized using the specific mean/std-dev requirements of the ImageNet-trained weights.

• The Output: The model generates a Softmax probability distribution across 1,000 classes, which is then decoded into human-readable labels with associated confidence scores.

• Deployment Note: The project originally used ResNet50, but was switched to MobileNetV2 (running on tensorflow-cpu) after hitting memory limits on free-tier hosting — a practical tradeoff between model size and deployability.

---

Known Limitations

• Trained on ImageNet, which covers common objects, animals, and everyday items — it has no concept of "person" or "face" as categories, so photos of people may return unrelated object labels. This is expected behavior for an ImageNet-trained classifier, not a bug.
• The backend is hosted on Render's free tier, which spins down after inactivity. The first request after idle time may take 30-60 seconds to respond.

---

👤 Author

Syed Ali Faraz - GitHub Profile

If you found this project useful, please give the repository a ⭐!