Hi Shahul, Thanks for reading it. Actually I cross checked in Tensor flow implemention, they have used first three layers as with stride 2 and rest with stride 1.
You can walkthrough the code with method name discriminator() where they have used 3 times as downsample in this link. From this link references, I try to illustrate with the example in the blog. https://www.tensorflow.org/tutorials/generative/pix2pix

Cheers!