Reshape PIL as 4d Tensor in Pytorch?

Question

I\'m playing with the PyTorch [DCGAN tutorial][1], and was wondering if I could replace the Gaussian noise the tutorial is using to initialize training, with my own custom i