What is the functionality of step_size in patching

[mesa666]

def seg_and_patch(source, save_dir, patch_save_dir, mask_save_dir, stitch_save_dir,
patch_size = 256, step_size = 128,
seg_params = {'seg_level': -1, 'sthresh': 8, 'mthresh': 7, 'close': 4, 'use_otsu': False,
'keep_ids': 'none', 'exclude_ids': 'none'},
filter_params = {'a_t':100, 'a_h': 16, 'max_n_holes':8},
vis_params = {'vis_level': -1, 'line_thickness': 500},
patch_params = {'use_padding': True, 'contour_fn': 'four_pt'},
patch_level = 0,
use_default_params = False,
seg = False, save_mask = True,
stitch= False,
patch = False, auto_skip=True, process_list = None):

What would happen (down the road when extracting features and training) if a step_size that is larger or smaller than the patch_size is used? As in the above example step_size = 1/2 patch_size.

Why was this feature added in the first place?

[his0car]

The primary purpose of variable step_size (smaller than patch_size) is for creating high-quality, smooth heatmaps rather than for training. In line 117 of create_heatmaps.py: step_size = tuple((np.array(patch_size) * (1 - patch_args.overlap)).astype(int)) the step size is computed and then passed along.