PyTorch: Inference on a single very large image using multiple GPUs?

Question

I want to perform inference (i.e. semantic segmentation) on a very large satellite image without splitting it into pieces. I have access to 4 GPUs (each having 15 GBs of memory)