Skip to content

README.md

Latest commit

 

History

History
75 lines (47 loc) · 2.33 KB

README.md

File metadata and controls

75 lines (47 loc) · 2.33 KB

Quickstart Example: YOLOv8 Instance Segmentation with ONNX Runtime

Overview

This example shows a minimal end-to-end luxonis-eval workflow that runs entirely on the host and does not require a Luxonis device.

The setup prepares everything needed for evaluation:

  • downloads a YOLOv8n-seg ONNX model from HubAI
  • downloads a small COCO 2017 sample through FiftyOne
  • parses the dataset into Luxonis Data Format (LDF) with luxonis_ml
  • runs instance segmentation evaluation with ONNX Runtime on CPU

Prerequisites

Install luxonis_eval from the repository root (luxonis_ml and other dependencies are pulled in automatically), and fiftyone for dataset handling:

# luxonis_eval install
pip install .

# install fiftyone
pip install fiftyone

The setup script also expects the following tools to be available in PATH:

  • wget
  • tar
  • fiftyone

Prepare the Model and Dataset

From the repository root, run:

bash examples/quickstart_inst_seg/setup_example.sh

This script:

  1. fetches a fresh signed download URL for the YOLOv8n-seg ONNX model from HubAI
  2. downloads and extracts the model into examples/quickstart_inst_seg/models/
  3. downloads small COCO 2017 splits with FiftyOne
  4. parses the dataset into LDF as coco-2017

To force the model to be downloaded again, run:

bash examples/quickstart_inst_seg/setup_example.sh --force

Run Evaluation

After the setup completes, run:

luxonis_eval eval --config configs/yolov8n_inst_seg_config.yaml

This evaluates the model on the prepared dataset using ONNX Runtime with the CPU execution provider.

What the Setup Script Does

Model download

The script calls the HubAI API to retrieve a fresh signed URL for the YOLOv8n-seg ONNX archive, downloads it, extracts yolov8n-seg.onnx, and stores it under examples/quickstart_inst_seg/models/. If the model already exists locally, the download step is skipped unless --force is passed.

Dataset download

The script downloads small COCO 2017 splits through FiftyOne to keep the example lightweight. By default, the downloaded data is stored under ~/fiftyone/coco-2017/.

LDF parsing

Once the dataset is available locally, the script runs luxonis_ml data parse and creates an LDF dataset named coco-2017, which is then used by the evaluation config.