#WinML – How to convert Tiny-YoloV3 model in CoreML format to Onnx and use it in a #Windows10 App

Windows 10 and YOLOV2 for Object Detection Series


In today’s post I’ll share some experience on how to download the latest version of Tiny-YoloV3 and use it in the UWP App that I created in previous posts. Important, the Tiny-YoloV2  model I’ve used in previous posts was in Onnx format, and it was downloaded from Azure AI Gallery.

We can download Tiny-YoloV3 from the official site, however I will work with a version that is already compiled in CoreML format, CoreML format is usually used in iOS apps (see References).

Well, to convert the model of CoreML To Onnx, we will use Visual Studio Tools For Ai. And the next set of software

Once all the software is installed, we can follow the step by step of [Ai Converting Models To ONNX] to convert our model. However, the road is not as simple as it seems. The 1st thing we can find are errors like the following

01 thanks Python

Problems with Python, in my case had several versions of Python installed, but the IDE had no marked as [Default]. From the panel [Pythin Environments] You can fix this

01 1 Python environments

Next Problem I need some configuration on my part, it was related to Prerequisites for conversion. It all starts with [Missing Package WinMLTools]




Missing package WinMLTools, please check details in output window.





Traceback (most recent call last):

  File “C:\Users\<Bruno>\AppData\Local\Microsoft\VisualStudio\15.0_e5344afb\Extensions\kzqekf1z.44v\RuntimeSDK\model\model_converter_cli.py”, line 76, in check_winmltools_installed

    import winmltools

ModuleNotFoundError: No module named ‘winmltools’

However After Of Investigate for a while, these are the 2 Packages I need to install

Microsoft ML Tool (winmltools)

pip3 install winmltools==

CoreML Tools

pip3 install “git+https://github.com/apple/coremltools@v0.8”

As always, 1st thing to do is update Python

03 upgrade pip.png

And now we can launch the conversion operation again

07 01 convert.png

And a few seconds later we already have our Tiny-YoloV3 in format Onnx

07 convert running

Because the model respects the Input/Output of the previous version, we only have to replace the file in our solution. I have added the new Onnx Just to have a little more control over the example.

08 Sol Onnx.png

As I have always updated the complete example in GitHub


Happy Coding!

Greetings @ Toronto

El Bruno




        1. Sorry, I pasted the wrong one.
          I’ve downloaded everything from https://github.com/AlexeyAB/darknet
          I’ve followed the steps from the section [How to train tiny-yolo], I’ll double check the file I’ve used and I’ll try to write my experience here.
          I’m not a ML tech guy, so at this point my experience is kind of messy to share 😀


Leave a comment

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.

%d bloggers like this: