Difference between revisions of "My model doesn't work"

From Emgu CV: OpenCV in .NET (C#, VB, C++ and more)
Jump to: navigation, search
(Initial write up.)
(No difference)

Revision as of 17:02, 3 April 2020

Why my model doesn't work

Try our demo project first

First, please try our demo project. Run it on your workstation or mobile device. In most of the case, our demo project should work as expected. If not, please send us the following information such that we can take a look:

  • What is the full package name. e.g. libemgutflite-android- or Emgu TF Lite v2.1 from Unity asset store.
  • Platform information. e.g. Windows, Mac, Android or iOS.
  • What is the CPU architecture. e.g. x86, x64, ARM or ARM64.
  • For mobile device, let us know if it is a simulator or device.
  • OS version. For Android, let us know the API level (e.g. Android 9) where the demo project is run.

Your demo works, I replaced with mine and it fails

  • Not all model are created the same. TF / TF Lite model that use different training method are not drop and replaced. Even models that use the same training script can have different type of input tensor and output tensor. e.g. quantized model has different input tensor type than the none-quantized model, even the training code is the same.
  • You need to make sure the TF / TF Lite model is working as expected. The best way to check is to load the model in python and test it against data, and make sure it produce the correct output. Or you can test it against Android device if you are using Android Studio.