SKU Detection Engine

A lightweight FastAPI server for Stock Keeping Unit (SKU) detection using quantized ONNX model (RetinaNet). This core detection engine provides the underlying ML inference capabilities for SKU detection workflows.

About

This detection engine detects products/SKUs in retail images using a trained RetinaNet model. The engine provides general SKU detection with a single detection class, suitable for:

• Retail inventory management
• Product recognition in shelf images
• Automated stock monitoring
• Package/product detection

Note: This engine supports general SKU detection only, not specific product classification. All detected items are detected as generic "products" (class_id: 1).

Architecture: This service acts as the core detection engine and can be integrated with higher-level APIs (like jake-detector-api) for enhanced business logic and workflow management.

Features

• Core SKU detection engine with quantized ONNX model
• Fast inference for real-time detection with optimized model size
• RESTful API endpoints for integration
• Base64 encoded result images with bounding boxes drawn around detected products
• Confidence scoring for detection quality assessment
• Direct integration ready for higher-level services

Installation

1. Install dependencies:

pip install -r requirements.txt

2. Ensure your model.onnx file is in the root directory

Usage

Start the detection engine:

python app.py

The detection engine will be available at http://localhost:8000

Documentation: Visit http://localhost:8000/docs for interactive Swagger UI

Integration: This engine can be integrated with higher-level APIs for enhanced workflow management.

API Endpoints

POST /detect/url

Detect SKUs from image URL

{
  "image_url": "https://example.com/shelf-image.jpg",
  "confidence_threshold": 0.3
}

POST /detect/upload

Detect SKUs from uploaded image file

• file: Image file (multipart/form-data)
• confidence_threshold: Confidence threshold (optional, default: 0.3)

GET /health

Health check endpoint

Response Format

The API returns detected SKUs with bounding box coordinates, confidence scores, and class IDs for product detection:

{
  "detections": [
    {
      "x1": 100,
      "y1": 200,
      "x2": 300,
      "y2": 400,
      "confidence": 0.85,
      "class_id": 1
    }
  ],
  "num_detections": 1,
  "image_base64": "base64-encoded-image-with-bounding-boxes"
}

Fields:

• detections: Array of detected SKUs with pixel coordinates
• num_detections: Total count of detected products
• image_base64: Original image with red bounding boxes drawn around detected SKUs
• class_id: Product/SKU detector for inventory management