Object Detection Service

Object detection service written in Rust with Onnx inference engine. Supports Blue Iris and Agent DVR.

TL;DR

Current features:

Feature	Windows x86_64	Linux x86_64
RT-DETR-V2 ONNX Models	🟢	🟢
Yolo 5 ONNX Models (including custom)	🟢	🟢
Run as a service	🟢	❌
Docker image	❌	🟢
CPU Inference	🟢	🟢
AMD GPU Inference	🟢	❌
Intel GPU Inference	🟢	❌
Nvidia GPU Inference	🟢	❌
Coral TPU Inference	❌	❌

Install on Windows with THE one mighty oneliner

 powershell -NoProfile -Command "curl 'https://github.com/xnorpx/blue-onyx/releases/latest/download/install_latest_blue_onyx.ps1' -o 'install_latest_blue_onyx.ps1'; Unblock-File '.\install_latest_blue_onyx.ps1'; powershell.exe -ExecutionPolicy Bypass -File '.\install_latest_blue_onyx.ps1'"

Install as service on Windows

Note: You need to run as administrator to register the service and change the install path and command line arguments for your setup.

sc.exe create blue_onyx_service binPath= "$env:USERPROFILE\.blue-onyx\blue_onyx_service.exe --port 32168" start= auto displayname= "Blue Onyx Service"
net start blue_onyx_service

Verify it is working by going to http://127.0.0.1:32168/

(If you don't want to run blue_onyx as a service you can just run blue_onyx.exe)

Docker container on Linux

docker pull ghcr.io/xnorpx/blue_onyx:latest
docker run -d -p 32168:32168 ghcr.io/xnorpx/blue_onyx:latest

I don't trust scripts I want to install myself

• Download latest release
• Unzip
• Run blue_onyx

Automatic Model Management

Blue Onyx automatically downloads and manages models for you:

• Default Model: rt-detrv2-s.onnx is used by default
• Auto-Download: Models are downloaded automatically on first use
• Multiple Model Types: RT-DETR v2 (general purpose) and YOLO5 (specialized)

Manual Model Download

# List all available models
blue_onyx --list-models

# Download all models
blue_onyx --download-model-path ./models --download-all-models

# Download only RT-DETR v2 models (recommended for general use)
blue_onyx --download-model-path ./models --download-rt-detr2

# Download only YOLO5 specialized models (IP cameras, delivery detection)
blue_onyx --download-model-path ./models --download-yolo5

Available Models

Model Type	Models	Use Case	Size
RT-DETR v2	rt-detrv2-s/ms/m/l/x	General object detection (80 COCO classes)	80MB - 400MB
YOLO5 Specialized	delivery, IPcam-animal, ipcam-bird, etc.	IP cameras, delivery detection	~25MB each

Quick Usage Examples

Basic Usage

# Start with default settings (auto-downloads rt-detrv2-s.onnx)
blue_onyx

# Start with specific model
blue_onyx --model ./models/rt-detrv2-l.onnx

# Start with specialized model for delivery detection
blue_onyx --model ./models/delivery.onnx --object-detection-model-type yolo5

Configuration

# Custom port and confidence threshold
blue_onyx --port 8080 --confidence_threshold 0.7

# Filter for specific objects only
blue_onyx --object_filter person,car,bicycle

# Force CPU usage (disable GPU)
blue_onyx --force_cpu

# Enable debug logging
blue_onyx --log_level Debug

Notes on Linux

If you run outside of docker you need to install OpenSSL 3

Performance Testing

Benchmark GPU

blue_onyx_benchmark --repeat 100 --save-stats-path .
Device Name,Version,Type,Platform,EndpointProvider,Images,Total [s],Min [ms],Max [ms],Average [ms],FPS
Intel(R) Iris(R) Xe Graphics,0.1.0,GPU,Windows,DML,100,14.3,116.8,168.3,143.2,7.0

Benchmark CPU

blue_onyx_benchmark --repeat 100 --save-stats-path . --force-cpu
Device Name,Version,Type,Platform,EndpointProvider,Images,Total [s],Min [ms],Max [ms],Average [ms],FPS
12th Gen Intel(R) Core(TM) i7-1265U,0.1.0,CPU,Windows,CPU,100,28.2,239.6,398.2,281.5,3.6

Testing

Test Service

blue_onyx

Then run in another terminal to do 100 requests with 100 ms interval:

test_blue_onyx --number-of-requests 100 --interval 100

Test Detection and Save Results

blue_onyx_benchmark --save-image-path .

API Usage

Detect Objects in Images

# Upload image file
curl -X POST -F "image=@test.jpg" http://localhost:32168/detect

# Detect from URL
curl -X POST -H "Content-Type: application/json" \
  -d '{"url": "https://example.com/image.jpg"}' \
  http://localhost:32168/detect

Web Interface

Open your browser and go to: http://localhost:32168/

Documentation

For detailed documentation, visit: Blue Onyx Documentation

• Getting Started - Quick start guide
• Models - Available models and usage
• Configuration - Detailed configuration options
• Windows Installation - Windows setup guide
• Linux Installation - Linux/Docker setup
• FAQ - Common questions and troubleshooting