PatchCore Damage Detection Service

Minimal Flask service for package damage detection using PatchCore anomaly detection. Features a modern web interface and RESTful API with CORS support for external integrations.

Quick Start

1. Create virtual environment:

uv venv

2. Activate virtual environment:

uv activate

3. Install dependencies:

uv pip install -r requirements.txt

4. Run service:

uv run app.py

5. Open browser:

http://127.0.0.1:5000

API Endpoints

GET /

Web interface for interactive damage detection.

POST /load_reference

Upload a normal (undamaged) reference image to build the memory bank.

Request:

• Content-Type: multipart/form-data
• Body: image (file)

Response:

{
  "status": "success",
  "message": "Memory bank built successfully",
  "patches": 784,
  "feature_dim": 1536
}

POST /infer

Upload a test image and receive the overlay PNG with anomaly heatmap.

Request:

• Content-Type: multipart/form-data
• Body: image (file)

Response:

• Content-Type: image/png
• Returns PNG image at original input resolution

POST /infer_json

Upload a test image and receive JSON with damage statistics and base64-encoded image.

Request:

• Content-Type: multipart/form-data
• Body: image (file)

Response:

{
  "status": "success",
  "damage_percentage": 12.34,
  "anomaly_detected": true,
  "threshold": 0.5,
  "image": "data:image/png;base64,..."
}

External API Usage

cURL Example

# Load reference image
curl -X POST http://127.0.0.1:5000/load_reference \
  -F "image=@normal_image.jpg"

# Get JSON response with stats
curl -X POST http://127.0.0.1:5000/infer_json \
  -F "image=@test_image.jpg"

# Get PNG image directly
curl -X POST http://127.0.0.1:5000/infer \
  -F "image=@test_image.jpg" \
  -o result.png

Python Example

import requests

# Load reference
with open('normal.jpg', 'rb') as f:
    response = requests.post(
        'http://127.0.0.1:5000/load_reference',
        files={'image': f}
    )
print(response.json())

# Get damage detection results
with open('test.jpg', 'rb') as f:
    response = requests.post(
        'http://127.0.0.1:5000/infer_json',
        files={'image': f}
    )
data = response.json()
print(f"Damage: {data['damage_percentage']}%")
print(f"Detected: {data['anomaly_detected']}")

Next.js/React Example

const formData = new FormData();
formData.append('image', file);

const response = await fetch('http://127.0.0.1:5000/infer_json', {
  method: 'POST',
  body: formData
});

const data = await response.json();
// Use data.image (base64), data.damage_percentage, data.anomaly_detected

Features

• Original Resolution: Output images match input image dimensions
• CORS Enabled: Ready for cross-origin requests from web applications
• Dual Output Formats: PNG image endpoint and JSON endpoint with statistics
• Real-time Detection: Fast inference with WideResNet-50-2 backbone
• Modern UI: Sleek web interface with drag-and-drop support

Configuration

Change device in app.py:

device = torch.device('cuda')  # For GPU

Change port in app.py:

app.run(host='127.0.0.1', port=8080, debug=False)

Notes

• Memory bank must be loaded via /load_reference before running inference
• Detection threshold is fixed at 0.5
• Anomaly is detected when damage percentage > 5%
• Output images are returned at original input resolution