Class Crop

diffplanner/datasets/pipelines/transforms.py:13–50

r"""Crop motion sequences. Args: crop_size (int): The size of the cropped motion sequence.

@PIPELINES.register_module()
class Crop(object):
    r"""Crop motion sequences.
    
    Args:
        crop_size (int): The size of the cropped motion sequence.
    """
    def __init__(self,
                 crop_size: Optional[Union[int, None]] = None):
        self.crop_size = crop_size
        assert self.crop_size is not None
        
    def __call__(self, results):
        motion = results['motion']
        length = len(motion)
        if length >= self.crop_size:
            idx = random.randint(0, length - self.crop_size)
            motion = motion[idx: idx + self.crop_size]
            results['motion_length'] = self.crop_size
        else:
            padding_length = self.crop_size - length
            D = motion.shape[1:]
            padding_zeros = np.zeros((padding_length, *D), dtype=np.float32)
            motion = np.concatenate([motion, padding_zeros], axis=0)
            results['motion_length'] = length
        assert len(motion) == self.crop_size
        results['motion'] = motion
        results['motion_shape'] = motion.shape
        if length >= self.crop_size:
            results['motion_mask'] = torch.ones(self.crop_size).numpy()
        else:
            results['motion_mask'] = torch.cat(
                (torch.ones(length), torch.zeros(self.crop_size - length))).numpy()
        return results
        
        
    def __repr__(self):
        repr_str = self.__class__.__name__ + f'(crop_size={self.crop_size})'
        return repr_str

@PIPELINES.register_module()
class RandomCrop(object):