Hi @isaacdevlugt, I adapted the quantum GAN demo for 64x64 images RGB. I used 13 qubits, 2 ancillary, 2 layers depth and 6 sub generators. The input to discriminator is 64x64x3, while the input of the generator is batch size x n.qubits. The training was really slow but it completed using alsoGPU and lightning.qubit or .gpu. Although after 800 epochs the output images are a black square with some coloured points and the discriminator loss reached a really low value.
So the problem is not how many epochs training but i think we need to modify something in the quantum generator. The output images have 4 dimensions(batch size, 64,64,3). Can you help me in order to make the model work for RGB images?
The code is the file ‘qgan’ on github.
Thank you