Hello. Im just started lwarning ML and have small question.
I have trained model, for now it is in h5 format. The model input is image, output is some numbers.
Question 1. Is it actually possible to run this type of model? Will i be able to give value and get value from model.
Question 2. Which format Unity need to run the model (pb, onnx, bytes,…)
Question 3. I should use barracuda right, instead of ml agents?
Any sources will be good.
Thanks.
Q1. Depends on the model and what operators are used, but potentially yes.
Q2. Using Barracuda in Unity you will need to generate an .onnx file.
Q3. Yes.
Take a look at GitHub - onnx/tensorflow-onnx: Convert TensorFlow, Keras, Tensorflow.js and Tflite models to ONNX for converting your model to onnx format.
Thanks you. i have succeeded.
Awesome! Did you get it to run in Unity?
yeap. it works perfectly.
Except for one thing. the worker type set to Auto crashes on my phone with the snapdragon video card.
So I changed to Csharp, and now it works. Didn’t test the other 5 types.
Out of curiosity, what is your phone model?
Can you also share the onnx file? That way we can investigate on our end
@Mantas-Puida Samsung galaxy note10+
@alexandreribard_unity No, i cant 
Sorry. It is a top secret
Okay, well, if you are able to reproduce the crash with a smaller version of the network that doesn’t reveal the full architecture or just the part that you think is causing the crash, then we can take a look at it.