Why does my tflite model predict well with Python interpreter, but very poorly when deployed in Android Studio?

Question

I have built a model using mobilenet for a specific training data set. When testing my model with a test set, the model generated in keras (model.h5) obtains an accuracy of appr