Closed apyskir closed 6 years ago
erode_selem_size
parameter is necessary in prepare_masks
dilate_selem_size
parameter is necessary in train/evaluate/predict
These params equal to 0 mean, that no erosion/dilation is performed. Normally you want to keep them the same.
If you use dilation in postprocessing, remember to change Y_COLUMNS
in pipeline_config.py to 'file_path_mask_eroded', so the model learns the eroded masks, not the full masks.