Create convolution layers share by D and G as Frontend.
D is the discriminator for identifying the real picture or fake picture.
G is generater which sample noise from uniform distribution and try to match with the latent variable.
Use mu and mean from convolution layers to generate images.
The loss function is combine reconstruction error, generater error, and discriminator error.
D learning rate has to be higher than G for better result. With small learning rate, G is able to generate realistic image
ProductionLine_imgae_detection.ipynb
Show the application and result of trained model applied on real manufatureing image and able to identified anomaly image using cross entorpy with accuracy 96%.
model_result.png
Show the sample image from generated model compare with real image.