Change TFLite Model for SSD

[ChrisMatchlock]

I had spent weeks on using my trained TFLite model for SSD.

The error it prompted is
"Exception: TensorFlowLite operation failed.
TensorFlowLite.Interpreter.ThrowIfError (TensorFlowLite.Interpreter+Status status) (at Packages/com.github.asus4.tflite/Runtime/Interpreter.cs:182)
TensorFlowLite.Interpreter.GetOutputTensorData (System.Int32 outputTensorIndex, System.Array outputTensorData) (at Packages/com.github.asus4.tflite/Runtime/Interpreter.cs:130)
TensorFlowLite.SSD.Invoke (UnityEngine.Texture inputTex) (at Assets/Samples/SSD/SSD.cs:37)
SsdSample.Update () (at Assets/Samples/SSD/SsdSample.cs:58)"

I tested the model and it is able to perform prediction....

I already used a qunatized tflite. What should I do to deal with this problem?

PS: The training setup I used from
"https://colab.research.google.com/github/EdjeElectronics/TensorFlow-Lite-Object-Detection-on-Android-and-Raspberry-Pi/blob/master/Train_TFLite2_Object_Detction_Model.ipynb"

See if you can tell me how to train the model in right format....

[ChrisMatchlock]

@asus4 I read the other thread. Seems like we are suggested to tweak the Unity code ourselve. Although I had a few years of Unity coding experience, I can not figure out where to tweak the code so that it could entertain a 320x320, 3, 0 model... Would you give me some hint? Thanks a lot.

[asus4]

@ChrisMatchlock, could you provide the tflite model here? There are several possible causes, including input and output formats and quantisation types, so it isn't easy to figure out without the model.

[ChrisMatchlock]

My pleasure to have your kind reply.

I uploaded the model into my goolge drive.

https://drive.google.com/drive/folders/17yZkAAXrJYO_k-stFy4gkAEaGBQyvFD8?usp=share_link

The model name is coin_quant.tflite. In the folder, I also attached the folder for me to test if the model worked or not.

Let me know if you need more information.
I guess that I should be close to make the application running right. I just need to fix this output size issue.

Looking forward to your comment. Thanks to you and your great Unity TFLite project! @asus4

[asus4]

@ChrisMatchlock It seems to be working!

coin.mp4

What I changed

• Used XNNPACK accelerator options because the model contains ops that are not supported on GPU delegate.
• Changed the output index because the output order has been changed since TF2
  

[ChrisMatchlock]

Oh. Thanks so much @asus4
After I changed the output order, my PC worked as you shown in the video. Great THANKS.

However, when I build to my mobile, it throw me the following and stop working.... Any idea?

2023-05-03 08:21:01.192 1014-1039 Unity com.asus4.tfliteunitysample W Interperter Warning: Restored original execution plan after delegate application failure.
TensorFlowLite.Interpreter:TfLiteInterpreterCreate(IntPtr, IntPtr)
TensorFlowLite.Interpreter:.ctor(Byte[], InterpreterOptions)
TensorFlowLite.BaseImagePredictor1:.ctor(String, Boolean) TensorFlowLite.SSD:.ctor(String) SsdSample:Start() 2023-05-03 08:21:01.209 1014-1039 Unity com.asus4.tfliteunitysample E Exception: Failed to create TensorFlowLite Interpreter at TensorFlowLite.Interpreter..ctor (System.Byte[] modelData, TensorFlowLite.InterpreterOptions options) [0x00000] in <00000000000000000000000000000000>:0 at TensorFlowLite.BaseImagePredictor1[T]..ctor (System.String modelPath, System.Boolean useGPU) [0x00000] in <00000000000000000000000000000000>:0
at TensorFlowLite.SSD..ctor (System.String modelPath) [0x00000] in <00000000000000000000000000000000>:0
at SsdSample.Start () [0x00000] in <00000000000000000000000000000000>:0
2023-05-03 08:21:01.210 1014-1039 Unity com.asus4.tfliteunitysample E NullReferenceException: Object reference not set to an instance of an object.
at SsdSample.Update () [0x00000] in <00000000000000000000000000000000>:0

[ChrisMatchlock]

It seemed that it cannot start the interpreter...

[ChrisMatchlock]

Oh. I think I just need to use the updated repo from you to set XNNPACK is alright. In PC, I just use NONE for the option.

Thanks so much for your kind assistant. Without @asus4's help, I am not able to find out the compatability output order issue.