🚗 Car Brand & Model Identification System

A production-ready computer vision application that identifies car brand and model from images.
The system is designed with a two-stage inference pipeline to improve robustness and real-world usability.

---

🔍 Project Overview

This project started as a car brand & model classifier and evolved into a more reliable system by adding a car / not-car gatekeeper and regional fine-tuning for Indian cars.

The application is deployed using Streamlit and supports real-time image uploads.

---

🧠 Architecture

1️⃣ Car / Not-Car Detection (Gatekeeper)

• Model: MobileNetV2
• Purpose: Filters out non-car images before classification
• Benefit: Reduces false positives and improves user experience

2️⃣ Brand & Model Classification

• Model: EfficientNet-based classifier
• Training data:
  • Stanford Cars Dataset
  • Additional Indian car images (fine-tuned)
• Output: Top-K predictions with confidence scores

---

🚀 Live Demo

👉 https://milind-pandya-car-brand-model-identifier.streamlit.app

🇮🇳 Indian Car Fine-Tuning (v2.2)

To improve performance on Indian roads, the classifier was fine-tuned with additional images of popular Indian car brands such as:

• Maruti Suzuki
• Tata
• Mahindra

Only 10–20 clean images per model were required, leveraging transfer learning and avoiding overfitting.

Note: Confidence scores may appear low due to a large number of classes; however, correct predictions consistently rank at the top.

---

🚀 Features

• ✅ Car / Not-Car detection
• ✅ Brand & model prediction
• ✅ Top-K predictions
• ✅ Streamlit-based UI
• ✅ Modular preprocessing & inference
• ✅ Ready for API / mobile extension

---

🛠️ Tech Stack

• Python 3.13
• TensorFlow / Keras
• MobileNetV2
• EfficientNet
• Streamlit
• NumPy, PIL
• icrawler (data collection)

---

📌 Version History

• v1.0 – Initial car brand & model classifier
• v2.1 – Added MobileNetV2 car/not-car gatekeeper
• v2.2 – Fine-tuned classifier with Indian car images