VGG-19 is a convolutional neural network that is 19 layers deep. You can load a pre-trained version of the network trained on more than a million images from the ImageNet database. 
       Using VGG19 pre-trained model as the model. And sending two input images to the same model but separate like good in model1 and bad in model2 and finally subtracting both model.

        Good: 

        Bad: 

          Detected defects & Major Defect.