Car Recognition Project

🚗 AI Car Model Identifier with Used Price Estimation – Built with Gradio & Hugging Face Spaces

 
3台の日本車（プリウス、フィット、ノート）が近未来の都市を走る様子。背景には「AI」の文字と回路図が輝き、車種判別AIプロジェクトのスタートを象徴するデザイン。

I recently developed a simple AI app that can identify the car model from an image and display the estimated average used car price.
This project uses Gradio for the user interface and is freely available on Hugging Face Spaces.

In this article, I’ll walk you through the key features, how it was built, and how you can try it yourself.

App Overview

This AI-powered web app can:

  • Let you upload a car image
  • Automatically predict the top 3 likely car models
  • For each candidate, it displays:
  • Model production period
  • Average used price (in JPY)
  • Description
  • Catalog page link
  • Results are shown in a visually appealing card-style UI, including confidence levels

The Gradio interface is intuitive and beginner-friendly—no installation required.

Try the Demo

👉 Click here to try the app

(Runs directly on Hugging Face Spaces – no setup needed!)

Technical Highlights

✅ Model

  • Architecture: PyTorch ResNet101 (with transfer learning)
  • Classes: ~100 Japanese car models
  • Accuracy: Over 96% on validation data

✅ Gradio-based UI

  • app.py defines the Gradio Interface
  • Outputs are rendered as HTML info cards with confidence levels
  • Threshold slider allows users to filter low-confidence predictions

✅ Extra Features

  • Japanese label mapping (via jp_name_map.json)
  • Average used prices and descriptions loaded from used_car_prices.json
  • Catalog links included in each card

Publishing with GitHub + Hugging Face Spaces

The app is hosted for free on Hugging Face Spaces, synced via a public GitHub repo.

🔐 Note: GitHub File Size Limit

  • Since the .pth model file exceeds 100MB, I used Git LFS (Large File Storage)
  • It’s essential to include a correct .gitattributes file when using LFS

🌐 Deployment Workflow

  1. Prepare a GitHub repo with your Gradio app
  2. Create a new Hugging Face Space and link it to your GitHub
  3. Push with Git LFS enabled – the app auto-builds and publishes

Future Plans

  • Add YOLO for automatic car detection before classification
  • Experiment with EfficientNet or ViT for higher accuracy
  • Improve mobile support and add multilingual UI

Summary

Building an app that combines AI, car recognition, and web interface has never been easier.
With just a few tools like Gradio and Hugging Face Spaces, anyone can deploy AI apps for the world to try.

If you’re a car enthusiast, AI hobbyist, or just curious about deep learning in action, feel free to give it a go!

👉 Try the live app:
https://huggingface.co/spaces/Wan-shu/kuruma-checker

ResNet
RELATED POST