Patch Size

Controls the size of image patches extracted during training.

The patch size determines the dimensions of square image regions used to train the network. During training, random patches of this size are extracted from your training images.

Smaller patches (e.g., 64x64) train faster and use less memory, but may miss larger structures in your images.

Larger patches (e.g., 256x256) capture more context and can better segment large structures, but require more GPU memory and train more slowly.