bkunters / tflite4unity Goto Github PK

Editor interface to import models via a visual interface.

• bazel build //tensorflow/lite:libtensorflowlite.so --config android_arm64 --cxxopt='--std=c++11' -c opt
• https://github.com/tensorflow/tensorflow/blob/90e6bdca1ffa37bec962034cc10d1655451fd6c3/tensorflow/lite/delegates/gpu/BUILD#L143
• https://zimenglyu.com/en/ml/android/tensorflow/2018/11/27/tflite-android-ndk-eng.html

hey guys,
I tried my model via your selfie segmentation sample but no matter what I do, it always gives the wrong output
these are model input and output dimensions:
input => type: Float32, dimensions: [1,256,512,3], height, width, channel
output => Int64, dimensions: [1,256,512,1], height, width, channel
(channel index is between 0-18 (index 11: person)

download the model with the below link:
https://drive.google.com/file/d/1cdSqIlg3ES7kRTZvmn8rxhYXCREIBSSC/view?usp=sharing

Thank you for doing this job.
I have a problem with the running of the project on Google pixel 2 Android 10, with Unity 2019.2.5 . I have used the development branch. Here is the error that I am getting from logcat.

AndroidJavaException: java.lang.IllegalArgumentException: Internal error: Failed to run on the given Interpreter: OpenCL library not loaded - dlopen failed: library "libOpenCL-pixel.so" not found Falling back to OpenGL TfLiteGpuDelegate Invoke: Write to buffer failed. Source data is larger than buffer.

Project is working properly on Google pixel 3 android 10 without changes.

Thanks in advance

Hi,

I have been searching desperately for a solution of using TFlite model on Unity to develop for Android device, I want to build an Android app with augmented reality (ARCore) and artificial intelligence (TFlite), and your plug-in seems to have the best hope so far.

I do have some experience in Unity, but I have ZERO experience in TensorFlow and machine learning. I understand that the output of TensorFlow is a pre-trained model in .tflite format. Assume I can get a model ( there's a bunch online already) for a simple ML purpose (say object detection).

My main question is: how do I set up the environment in Unity to use this .tflite model? For example, a pretraining object tracking model should give me output as the target names and the bounding boxes. What should I do in Unity to make the use of this model?

Please bear with me if this sounds extremely stupid, but I am so desperately in need of this information, and I cannot find it anywhere else.

PS: the TFLite Android link in your post is broken.

Hello,

Thank you for your project. I succesfully build it and run it on Galaxy S10. And the frame rate was really good !
I have 2 questions concerning the development pipeline and the usage of ARCore. To say it clearly, i know nothing about MediaPipe and my knowledge in Android development is pretty low. So sorry in advance if the answered are triviale :)

• Is there a way to have the preview in the Unity editor window (with TFLite running)
• Is there any strong dependency with ARCore, or it's simple to remove it from your example project ?

Regards,
--Selim

First of all thank you for your great effort. I don't get the hand tracking.
Unity 2019.3.0b5
JDK 1.8.0_152
Google Pixel XL 8.1.0
I hope you can help me.
This is the message captured in the debug console:

(Filename: ./Runtime/Export/Debug/Debug.bindings.h Line: 35)
2019-10-16 08:43:20.339 7519-7535/? E/Unity: AndroidJavaException: java.lang.NullPointerException: Attempt to invoke virtual method 'void org.tensorflow.lite.Interpreter.runForMultipleInputsOutputs(java.lang.Object[], java.util.Map)' on a null object reference
java.lang.NullPointerException: Attempt to invoke virtual method 'void org.tensorflow.lite.Interpreter.runForMultipleInputsOutputs(java.lang.Object[], java.util.Map)' on a null object reference
at com.bora.tflitehandtracking.HandTrackingInference.RunInference(HandTrackingInference.java:113)
at com.unity3d.player.UnityPlayer.nativeRender(Native Method)
at com.unity3d.player.UnityPlayer.access$300(Unknown Source:0)
at com.unity3d.player.UnityPlayer$e$1.handleMessage(Unknown Source:88)
at android.os.Handler.dispatchMessage(Handler.java:102)
at android.os.Looper.loop(Looper.java:164)
at com.unity3d.player.UnityPlayer$e.run(Unknown Source:20)
at UnityEngine.AndroidJNISafe.CheckException () [0x00000] in <00000000000000000000000000000000>:0
at UnityEngine.AndroidJNISafe.CallObjectMethod (Sys