ARIES Multi-Channel Demo

ARIES Multi-Channel Demo is a sample application for running multi-channel video inference workflows on MLA100-based systems.

1) Overview

This project supports:

• Linux and Windows build/run workflow
• MLA100-based hardware inference
• Multiple feeder types (camera/RTSP/YouTube/video file)
• Multiple model types (SSD / STYLENET / FACE / POSE / OBJECT / SEGMENTATION)
• Layout and worker configuration through YAML files

2) Prerequisites (single source)

Linux

• Ubuntu 20.04 LTS or newer
• Git and Git LFS
• build-essential, cmake
• libopencv-dev
• Mobilint runtime packages:
  • mobilint-aries-driver
  • mobilint-qb-runtime
  • mobilint-cli
• Optional: python3, pip, and yt-dlp (required for YouTube feeder)

Windows

• Windows 10/11
• Visual Studio 2022
• CMake
• OpenCV 4.x (installed in your environment)
• MLA100 runtime and driver support

3) Linux setup and run flow

1. Install required system packages:

sudo apt update
sudo apt install -y libopencv-dev mobilint-qb-runtime python3-pip

2. Get LFS assets (.mxq and sample .mp4):

git lfs install
git lfs pull

3. Build/update and register desktop shortcut:

./update.sh

4. Run the demo:

./run.sh

or directly:

cd build && ./src/demo/demo

If run.sh prints Demo binary not found, run ./update.sh first.

4) Build options

Build manually (Linux)

mkdir build
cd build
cmake [-DQBRUNTIME_PATH=<path>] ..
make -j
./src/demo/demo

cmake and other build dependencies are required. If CMake cannot be installed via package manager, install it from kitware:

https://apt.kitware.com/

Packaging

./package/package.sh aries2-v4 aries2

5) Required model files in this repository

The current example configuration uses the following mxq files:

• ../mxq/ssd.mxq
• ../mxq/yolo11s-face.mxq
• ../mxq/yolo26s-seg.mxq
• ../mxq/yolo26s-pose.mxq
• ../mxq/yolo26s.mxq

All .mxq and sample .mp4 assets are managed by Git LFS.

6) Feeder setting (rc/FeederSetting.yaml)

FEEDER_TYPE: { CAMERA | IPCAMERA | YOUTUBE | VIDEO }

- feeder_type: FEEDER_TYPE
  src_path:
    - <path or url>
- feeder_type: ...

feeder_type	src_path	Description
CAMERA	Camera index	Enter camera index as a number
IPCAMERA	RTSP URL	Enter RTSP URL
YOUTUBE	YouTube URLs	Enter one or more URLs
VIDEO	Video file paths	Enter one or more local video file paths

7) Model setting (rc/ModelSetting.yaml)

MODEL_TYPE: { SSD | STYLENET | FACE | POSE | OBJECT | SEGMENTATION }

CLUSTER: { Cluster0 | Cluster1 }

CORE: { Core0 | Core1 | Core2 | Core3 }

subconfig is required for FACE, POSE, OBJECT, and SEGMENTATION.

- model_type: MODEL_TYPE
  mxq_path: <path to mxq>
  dev_no: <board index>
  core_id:
    - cluster: CLUSTER
      core: CORE
  # optional: num_core: INTEGER
- ...

For SSD and STYLENET, subconfig is not required.

8) Layout setting (rc/LayoutSetting*.yaml)

image_layout:
  - path: /path/to/image
    roi: [x, y, w, h]
feeder_layout:
  - [x, y, w, h]
worker_layout:
  - {feeder_index: <index>, model_index: <index>, roi: [x, y, w, h]}

9) Keyboard shortcuts

Key	Action
D, d	Display FPS
T, t	Display Time
M, m	Maximize Screen
C, c	Clear All Workers
F, f	Fill All Workers
Q, q	Quit
Esc	Quit

Windows

OpenCV installation

Download OpenCV 4.x installer from official releases:

https://github.com/opencv/opencv/releases

Build commands

mkdir build
cd build
cmake -G "Visual Studio 17 2022" -DCMAKE_PREFIX_PATH="C:\path\to\opencv\build" -DQBRUNTIME_PATH="C:\path\to\qbruntime" ..

Then open the generated .sln file and build.

License

This project is released under the MIT License. See LICENSE.

Third-Party Notices

This project depends on external components such as OpenCV, Mobilint runtime/CLI packages, and optional yt-dlp.
See NOTICE for dependency licenses and notices.