I have trained a YOLOv8 model and successfully converted the resulting .pt file into a TensorFlow Lite .tflite model using the YOLO TFLite converter. I intend to integrate this model into a Flutter application.

However, I noticed that the output signature of my custom TFLite model is different from the output signature of the open-source TFLite models documented in the official TensorFlow Lite documentation.

Comparison of Output signature

Open-Source Model:
Name: TFLite_Detection_PostProcess
Shape: [1, 10, 4]
Data Type: float32
Name: TFLite_Detection_PostProcess:1
Shape: [1, 10]
Data Type: float32
Name: TFLite_Detection_PostProcess:2
Shape: [1, 10]
Data Type: float32
Name: TFLite_Detection_PostProcess:3
Shape: [1]
Data Type: float32

my Model:
Name: Identity
Shape: [1, 11, 8400]
Data Type: float32

What steps can I take to modify the output signature of my TFLite model to match the format used by the officially supported TFLite models? Specifically, I am looking for guidance on how to align my model’s output with the expected format for seamless integration into my Flutter app.

I read documents and try to modify pt, but I still not solve it.