ehsanulhaque92/AuraScanAI

AuraScanAI - Vehicle Damage Assessment API

This repository contains the complete backend service for the AuraScanAI project, a sophisticated AI-powered system for analyzing vehicle damage from images. The API is built with Flask and serves a custom-trained, multi-task Vision Transformer (ViT) model capable of identifying damage areas and assessing their severity.

The live API is deployed as a Docker container on Hugging Face Spaces.

🚀 Key Features

• —AI-Powered Analysis: Leverages a state-of-the-art Vision Transformer (ViT) model fine-tuned on over 15,000 images of vehicle damage.
• —Multi-Task Learning: The model simultaneously predicts:
• —The location of the primary damage area (bounding box).
• —The overall severity of the damage (minor, moderate, severe).
• —Business Logic Engine: Includes a post-processing layer to translate AI outputs into actionable business insights, including a realistic estimated repair cost range.
• —Scalable Architecture: Built with a professional, singleton pattern to ensure the large AI model is loaded only once, providing fast and efficient inference.
• —Containerized & Deployable: Fully containerized with Docker and configured for seamless deployment on cloud platforms like Hugging Face Spaces.

🛠️ Technology Stack

• —AI Framework: PyTorch
• —Vision Model Library: timm (PyTorch Image Models)
• —API Framework: Flask
• —WSGI Server: Gunicorn
• —Containerization: Docker
• —Cloud Deployment: Hugging Face Spaces

⚙️ API Endpoints

The server provides two main endpoints:

1. Health Check

A simple endpoint to verify that the server is running and responsive.

• —Endpoint: /ping
• —Method: GET
• —Success Response (200):

    {
      "message": "Server is alive!",
      "status": "ok"
    }

2. Damage Analysis

The core endpoint for analyzing an image.

• —Endpoint: /analyze
• —Method: POST
• —Request Body: multipart/form-data with a single field:
• —file: The vehicle image file (.jpg, .png, etc.).
• —Success Response (200): A detailed JSON object containing the full analysis.

    {
      "success": true,
      "totalDamages": 1,
      "overallSeverity": "severe",
      "confidence": "0.66",
      "costRange": {
        "min": 800,
        "max": 2500
      },
      "damages": [
        {
          "id": "dmg-1",
          "type": "Primary Damage Area",
          "location": "Detected by AI",
          "severity": "severe",
          "estimatedCost": { "min": 800, "max": 2500 },
          "coordinates": [
            [ 423.06, 364.49, 1229.91, 859.14 ]
          ]
        }
      ]
    }

• —Error Response (4xx/5xx):

    {
      "success": false,
      "error": "Descriptive error message."
    }

📜 MVP Approach & Future Roadmap

This project serves as a powerful Proof of Concept (MVP), demonstrating a complete end-to-end pipeline for AI-powered vehicle damage assessment.

Current Capability (MVP): The current AI model is an Image Assessment Model, designed to identify the single most prominent damage area in an image. It provides a holistic analysis, including an overall severity classification, estimated repair cost, and a bounding box for the primary damage region. This successfully proves the core technology is viable.

Future Roadmap: The next phase of this project will involve evolving the AI core into a full Multi-Object Detector (e.g., using a DETR or YOLO architecture). This will enable the system to:

• —Identify and draw bounding boxes for multiple, distinct damages in a single image.
• —Provide a detailed breakdown and cost estimate for each individual damage in the Damage Ledger.