yolo-python-onnx

YOLOv8 Segmentation with Python

This directory illustrates how to run YOLOv8 Segmentation through the MAX Engine. Specifically, this example uses Ultralytics YOLO with opencv to segment images from the webcam.

Quickstart

This example will change depending on if you have a webcam or not. If you have a webcam, the example will capture the webcam and display a window running live segmentation.

If you do not have a webcam, the example will segment a downloaded video file and generate output.mp4.

Magic instructions

If you have magic, you can run the following command:

magic run bash run.sh

Conda instructions

Create a Conda environment, activate that environment, and install the requirements:

# Create a Conda environment if you don't have one
conda create -n max-repo
# Update the environment with the environment.yml file
conda env update -n max-repo -f environment.yml --prune
# Run the example
conda run -n max-repo --live-stream bash run.sh

Note: GUI Dependencies

The webcam version of this example depends on opencv-python and its ability to render GUIs. The dependencies for this are not always installed on linux.

Downloading these dependencies is distro dependent. On Ubuntu, it should be:

apt install -y libgl1

Performance Comparison

To get an idea of the performance difference between the ONNX Runtime and MAX Engine, you can run the webcam script with/without the --onnx flag.

The model execution time is printed as part of the window title. Note, it can take a bit for the execution time to stabilize.

MAX Engine:

./segment.py webcam

ONNX Runtime:

./segment.py webcam --onnx

Files

• download-model.py: Downloads YOLOv8n-seg from ultralytics and exports it as ONNX.

• segment.py: Example program that runs the full YOLO segmentation pipeline through the MAX Engine on images from the webcam or video files. Use --help to see the various commands.