DNN model changes after post training quantization

Question

I\'ve been following this tutorial for developing a yolo network with keras.

The non-quantization model looks like this : last layers of the model

An output li