Why is the mobilenetv2 network model trained under tensorflow2 changed from saved_ When the model format is converted to tflite model, the test recognition accuracy becomes 0?

I trained the mobilenetv2 network model under tensorflow 2 and saved the model as saved_ Model format, the test recognition result is correct, but the test result is completely wrong when converting it to tflite model. What is the reason?

Hi @mengmeng,

Which device are you running the model on? Could you also share your Edge Impulse project ID?