Training a conditional GAN model

So, i am implementing a conditional GAN from scratch same as pix to pix to convert satellite image to maps… I collected the dataset (1000 satellite and maps images) manually and trained the model for 100 epochs., The output was quite good. Then, again I trained that model for next 100 epochs on that same dataset and found the the model recognized the blocks better for training images but was almost same for test satellite images…And I collected more 1500 satellite and maps images and tried to implement incremental training by making a new dataset of 1000 images (750 new image and 250 old images) and trained the model for 80 epochs… The output for this model to seen and unseen data was blank… What steps should I take after this? I am quite stuck and don’t know what to do.?