Hello, I am new to GANs. I aim at building a GAN model to start generating Indoor scenes (of size 128x128 or 256x256). I have a quite big dataset (more that 30k images).
I gave a try with DCGAN in PyTorch framework, using this implementation (https://github.com/eriklindernoren/PyTorch-GAN/blob/master/implementations/dcgan/dcgan.py) with BCE loss, but didn’t manage to generate even a structure “close” to the desired target distribution.
Due to the fact that DCGAN is an old architecture (noticed that it uses upsampling instead of transposed convolutions, and Is used mostly for image sizes of 32x32). Could someone please give me some advice on a good starting point for my model (any PyTorch repo with GAN architecture and loss) so that I can further build my implementation?*