The testing accuracy is not based on testing set, but rather manipulated result

[WayneDW]

In the testing set, basically what you are doing is first to find what images are not fake and then find the accuracy when they are classified as real.

So the accuracy can be really high when there is only a small number of images classified as real, which is kind of like cheating.