EdgeFlow

EdgeFlow is a repository for ONNX inference demos and edge deployment experiments.

The current project entry point is:

• mobilenetv2-demo/: MobileNetV2 ONNX Runtime demo on macOS, including model analysis, image inference, and CoreML/CPU benchmark scripts

Documentation

• docs/mobilenetv2-edge-deploy-optimization.md: technical note covering MobileNetV2 ONNX export, graph optimization, and multi-backend benchmarking

Quick Start

Enter the demo directory and install dependencies:

cd mobilenetv2-demo
python3 -m venv .venv
source .venv/bin/activate
python -m pip install -r requirements.txt

Download the release assets before running the demo:

python scripts/download_release_assets.py \
  --base-url https://github.com/kaysono0/EdgeFlow/releases/download/v0.1.0

Then run:

python scripts/analyze_model.py
python scripts/mobilenetv2_onnx_demo.py
python scripts/benchmark.py

Release Assets

Large ONNX binaries are distributed through GitHub Releases instead of git history.

Expected assets:

• mobilenetv2.onnx
• mobilenetv2.onnx.data
• mobilenetv2_sim.onnx
• resnet18.onnx
• resnet18.onnx.data

Repository Layout

.
├── README.md
├── docs
│   └── mobilenetv2-edge-deploy-optimization.md
├── mobilenetv2-demo
│   ├── README.md
│   ├── models
│   ├── release-assets.json
│   ├── requirements.txt
│   ├── requirements-export.txt
│   └── scripts
└── .gitignore

See mobilenetv2-demo/README.md for project-specific details.