Poker One Shot Detector App

[suyashcjoshi]

Hello,

I just saw the WWDC 2019 video and found out about this amazing project. I'm looking for the Poker app that was demo'ed using One Shot Object Detector api but can't find it anywhere. Is that code provided anywhere?

Ref: https://developer.apple.com/videos/play/wwdc2019/420/

Thanks!

[suyashcjoshi]

@shantanuchhabra was it shared?

[shantanuchhabra]

Hi @SuyashJoshi ,

Thanks for checking out our WWDC 19 video! Unfortunately, the code for the Three Card Poker App isn't available, but our user guide should help you get started. One approach could be to click an image of each playing card (52 images in total), construct an SFrame, and pass that SFrame in to turicreate.one_shot_object_detector.create to get a One Shot Object Detector model that can detect playing cards!

Let us know if you have any follow-up questions or if we can help in any way!

[suyashcjoshi]

Got it thanks, will try that. One more question, can I export model in another format such as for TensorFlow Lite or ONNX from Turi as I would love to deploy not only on iOS but also on other devices including IOT. Thanks again.

[shantanuchhabra]

You can export models to the Core ML format using model.export_coreml.

Remainder of my answer is sourced from tf-coreml/tf-coreml#316 (comment):
Once you have a Core ML model, you can try converting to ONNX using onnxmltools and run using the ONNX runtime. Further, you can try converting to Tensorflow using https://github.com/onnx/onnx-tensorflow.