[Beginner] DCGAN output uint8 => int16

1

Hello!

I’m trying to generate new terrain data by using a custom datasets containing TIF files from SRTM Mission which contains Earth elevation data in int16.

I’ve followed the PyTorch DCGAN Tutorial and it works! I get an output of 64x64 images.

My problem is, how can I indicate to my GAN that my input images are in int16 and how to get an output in int16 too.

Thanks for your answers!
Regards

2

Modules usually accept floating point dtypes, so you won’t be able to convert the layers to torch.int16:

lin = nn.Linear(1, 1).to('cuda').to(torch.int16)
> TypeError: nn.Module.to only accepts floating point dtypes, but got desired dtype=torch.int16

If you need a specific range in the output images (based on int16), you could try to clamp the output.