ONNX support for latest versions of Core ML #590
Description
Question
According to the README: “Code for ONNX to Core ML conversion is now available through coremltools python package and coremltools.converters.onnx.convert is the only supported API for conversion. “, indicating that coremltools had first-party support for ONNX conversion back in 2020. However, it seems like this is no longer the case, Apple states that: “The Unified Conversion API supports only TensorFlow and PyTorch neural networks. For converting models using multi-backend Keras, Caffe, ONNX, or other neural networks, use the conversion APIs specific to those packages.”. Therefore, the first-party support for ONNX has been dropped and the responsibility is back to ONNX to develop support. However, this repository seems like it is no longer being maintained. It is unclear to me why this is the case, and why Apple has dropped first-party support? My hypothesis is that you are able to convert ONNX models to PyTorch first and then to Core ML, but this seems like a hassle which is easily solved by a library such as this one.
So my questions are:
- Why did Apple drop first-party support for ONNX?
- Is this repository still being maintained?
- If so, why is this repository no longer being maintained, keeping in mind that ONNX aims to be the go-to platform for interopability.