Advantech-YOLO-Vision-Applications Container

A professional toolkit for deploying optimized YOLOv8 vision applications container on Advantech edge AI devices with hardware acceleration.

Overview

This repository provides a streamlined solution for running YOLOv8 computer vision applications container on Advantech edge AI hardware. The Container includes toolkit which automatically detects your device capabilities and sets up an optimized environment for computer vision tasks with full hardware acceleration support.

Designed specifically for Advantech edge AI devices based on NVIDIA Jetson platforms, this toolkit enables rapid deployment of object detection, instance segmentation, and classification applications with minimal configuration required.

Features

• Complete Docker Environment: Pre-configured container with all necessary hardware acceleration settings
• Optimized Model Management: Tools for downloading and converting YOLOv8 models to accelerated formats
• Hardware Acceleration Support: Full integration with NVIDIA CUDA, TensorRT, and GStreamer
• X11 Display Support: Seamless visualization of model outputs directly from the container
• Multiple Vision Applications: Ready-to-use applications for object detection, segmentation, and classification

Applications Included

Object Detection

• Real-time object detection using YOLOv8
• Support for 80+ COCO dataset classes
• Configurable confidence thresholds and post-processing

Instance Segmentation

• Pixel-level object segmentation for precise boundary detection
• Multi-class segmentation capabilities
• Visualization tools for segmentation masks

Object Classification

• High-accuracy image classification
• Support for custom classification tasks
• Class confidence visualization

Quick Start

1. Clone this repository:

git clone https://github.com/Advantech-EdgeSync-Containers/Advantech-YOLO-Vision-Applications.git
cd Advantech-YOLO-Vision-Applications

2. Start the container environment:

chmod +x build.sh
./build.sh

3. The Docker container will launch with all necessary hardware acceleration. You can access the applications as described in the Usage sections below.

Utility Usage

Model Loading Utility

The advantech-coe-model-load.py utility helps download optimized YOLOv8 models for your Advantech device:

python3 src/advantech-coe-model-load.py 

Parameters:

• task: Choose from 'detection', 'segmentation', or 'classification' (default: detection)
• size: Model size, use 'n' for nano or 's' for small (default: based on device)
• dir: Directory to save models (default: current directory)

Examples:

# Download a YOLOv8n detection model
python3 src/advantech-coe-model-load.py 

# Download a YOLOv8s segmentation model
python3 src/advantech-coe-model-load.py 

Model Export Utility

The advantech-coe-model-export.py utility converts YOLOv8 models to optimized formats for edge deployment:

python3 src/advantech-coe-model-export.py

Parameters:

• task: Choose from 'detection', 'segmentation', or 'classification'
• size: Model size ('n' or 's' recommended)
• format: Export format (onnx, engine, torchscript)
• device: Device for optimization (cpu or 0 for GPU)
• half: Enable half precision (FP16) for faster inference

Examples:

# Export YOLOv8n to ONNX format
python3 src/advantech-coe-model-export.py 

# Export YOLOv8s segmentation model to TensorRT engine with half precision
python3 src/advantech-coe-model-export.py 

Application Usage

YOLOv8 Vision Application

The main advantech-yolo.py application offers a complete solution for running YOLOv8 models:

python3 src/advantech-yolo.py 

Results:

Object Detection

Instance Segmentation

Classification

Parameters:

• --input: Path to video file, image, or camera device ID (0 for primary camera)
• --model: Path to YOLOv8 model file or model name (e.g., 'yolov8n.pt')
• --task: Task type ('detect', 'segment', 'classify')
• --conf: Confidence threshold (default: 0.25)
• --show: Display results in real-time window
• --save: Save results to output directory
• --device: Device to run inference (cpu or 0 for GPU)

Examples:

# Run object detection on a test video
python3 src/advantech-yolo.py --input data/test.mp4 --task detect --model yolov8n.pt --show

# Run instance segmentation on camera feed
python3 src/advantech-yolo.py --input 0 --task segment --model yolov8n-seg.pt --conf 0.3 --show

# Run classification on an image
python3 src/advantech-yolo.py --input data/image.jpg --task classify --model yolov8n-cls.pt --save

Step-by-Step Usage Guide

1. Set Up Environment:

   • Start the Docker container using ./build.sh
   • This initializes all hardware acceleration settings and dependencies

2. Download Models:

   • Use advantech-coe-model-load.py to download appropriate YOLOv8 models
   • Choose model size based on your device capability ('n' or 's' recommended)

3. Convert Models (Optional):

   • Use advantech-coe-model-export.py to convert to optimized formats
   • TensorRT format (engine) provides the best performance on Advantech devices

4. Run Applications:

   • Use advantech-yolo.py with appropriate parameters for your use case
   • Configure input source, model, and task type as needed
   • Enable visualization with --show parameter

5. Analyze Results:

   • View real-time results on screen or examine saved outputs
   • Output includes annotations, bounding boxes, segments, or classifications based on task

Directory Structure

.
├── data/               # Sample test data (includes test.mp4)
├── src/                # Source code for applications
│   ├── advantech-coe-model-export.py  # Model export utility
│   ├── advantech-coe-model-load.py    # Model download utility
│   └── advantech-yolo.py              # Main YOLO application
├── docker-compose.yml  # Docker Compose configuration
├── build.sh            # Build script
├── LICENSE             # License information
└── README.md           # This file

Performance Recommendations

For optimal performance on Advantech edge AI devices:

Task	Recommended Model	Optimal Format
Object Detection	YOLOv8n/s	TensorRT Engine
Instance Segmentation	YOLOv8n-seg/s-seg	TensorRT Engine
Classification	YOLOv8n-cls/s-cls	TensorRT Engine

The toolkit automatically selects appropriate configurations based on your device, with preference for smaller and more efficient models (n and s variants) to ensure real-time performance.

Limitations

The current version of the toolkit has the following limitations:

1. Model Size Constraints: Only supports 'n' and 's' model variants to maintain real-time performance on edge devices. Larger models may exceed memory or computational capabilities of some devices.

2. Pre-trained Models Only: Currently limited to pre-trained YOLOv8 models. Custom model training requires external workflows.

3. Resolution Limits: Performance degrades with very high-resolution inputs (>1080p). For best results, use input at 640×640 or 1280×720 resolution.

4. Network Dependency: Initial model downloading requires internet connectivity.

5. X11 Display Requirements: Visualization features require X11 forwarding to be properly configured.

6. Fixed Detection Classes: Models use pre-trained COCO classes and cannot be dynamically changed without retraining.

7. Single Stream Processing: Currently supports processing one input stream at a time.

Future Work

The following enhancements are planned for future releases:

• Vision Language Models (VLM): Integration with multimodal models for combining visual and text understanding
• Video Analytics: Adding object counting, dwell time analysis, trajectory tracking, and heat maps
• Multi-stream Processing: Support for processing multiple camera streams simultaneously
• REST API Interface: HTTP API for remote inference triggering and results retrieval
• Anomaly Detection: Algorithms for identifying unusual patterns and behaviors
• Automated Model Optimization: Dynamic model adaptation based on deployment conditions
• Scene Understanding: Moving from object detection to comprehensive scene interpretation
• Multi-sensor Fusion: Integration with non-visual sensors for enhanced perception
• Autonomous Decision Making: Enabling edge devices to make inferences that trigger actions

Use Cases

This toolkit is ideal for:

• Industrial Quality Inspection: Detect defects and inspect parts with instance segmentation
• Smart Retail: Product recognition, customer behavior analysis
• Smart Cities: Traffic monitoring, crowd analysis, object tracking
• Security & Surveillance: Perimeter monitoring, intrusion detection
• Agriculture: Crop monitoring, livestock tracking
• Healthcare: Medical image analysis, equipment tracking
• Robotics: Environmental perception, object manipulation guidance

License

Copyright (c) 2025 Advantech Corporation. All rights reserved.

This software is provided by Advantech Corporation "as is" and any express or implied warranties, including, but not limited to, the implied warranties of merchantability and fitness for a particular purpose are disclaimed.

For complete license details, see the LICENSE file.

Acknowledgments

• Ultralytics: For the YOLOv8 framework that powers this toolkit.

  # Required specific versions:
  pip install ultralytics==8.0.43 --no-deps
  pip install ultralytics-thop==2.0.14 --no-deps

• NVIDIA: For CUDA, TensorRT, and other acceleration libraries that enable optimal performance on Advantech edge AI devices.