I trained a keras model inspired by the Inception Model on Tensorflow backend. The problem is, the ouput is always the same, for differents images I tested.
However, model.evaluate give me a high accuracy percentage, so, the model seems to work.
Have you an idea ? Thanks!