It might not affect training as much as you think - you could also modify your loss function so it doesn’t penalise false positives as much as false negatives

Make sure you’re putting the model into evaluation mode during validation and zero your optimiser.

Just train on the cloud, don’t stress about your laptop, it won’t ever have the GPU for DL, unless you want to shell out for a Lambda machine.

Downsize your inputs. Or try a free GPU on paperspace.