onnx-coreml-conversion

This repository demonstrates how to convert an ONNX model (.onnx) into a CoreML model (.mlmodel) using Docker and a pinned environment.

---

Prerequisites

On macOS, make sure Docker is installed and running. The easiest way is:

brew install --cask docker

Then start Docker Desktop once so the Docker daemon is available.

---

One-Liner Conversion

Run this from the directory containing your ONNX file (replace):

docker run --platform=linux/amd64 --rm -v $PWD:/work python:3.8 bash -c "\
  pip install coremltools==4.1 onnx==1.7.0 protobuf==3.20.3 && \
  python -c \"import coremltools as ct; mlmodel = ct.converters.onnx.convert(model='/work/model.onnx'); mlmodel.save('/work/model.mlmodel')\" \
"

---

Output

After running, you will have:

• model.mlmodel in the same directory as your ONNX file
• A ready-to-use CoreML model for Xcode and Vision

---

Why This Works

• Uses Python 3.8 with --platform=linux/amd64 to download prebuilt wheels
• Pins CoreMLTools 4.1, ONNX 1.7.0, and Protobuf 3.20.3 (a stable, compatible combination)
• Runs fully inside Docker, keeping your local environment clean