I only made this code to keep the interface (NervetUPT, ultrasound_nerve_brachial_plexus)

(__call__, load_instance_by_id, and order of tuple),

I still recommend using tfds to create the datasets

class image_segmentation.datasets.segmentation.oxford_iiit_pet.OxfordIiitPet(split=[70, 15, 15], seed: int = 42, one_hot: bool = True)[source]
load_instance_by_id(id_img)[source]
to_one_hot(mask)[source]