Added netg.eval() for inference mode, to fix the behaviour of Batchnorm

[KMvanHespen]

Added netg.eval() for inference mode, to fix the behaviour of Batchnorm. Without this line, batch statistics are used during inference. Instead, we want to use running statistics during inference.

[hsekki]

Hi @KMvanHespen , @samet-akcay , I'm trying to use ganomaly as classifier to classify an image as normal(0) or abnormal(1). But I'm having trouble implementing it cause I think it is not straightforward. I'm done training with normal samples and tested with normal and abnormal. Could it be used also in inference of one image and return 0 if normal and 1 if it is abnormal? It would be a great help coz I feel stuck. Thanks in advance.