def __call__(
self, input: torch.tensor, instances: torch.tensor,
bounding_boxes: torch.tensor
) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
"""
Flipping augmentation (only horizontal)
:param image: (torch.Tensor) Input image of shape [channels, height, width]
:param instances: (torch.Tenor) Instances segmentation maps of shape [instances, height, width]
:param bounding_boxes: (torch.Tensor) Bounding boxes of shape [instances, 4 (x1, y1, x2, y2)]
:return: (Tuple[torch.Tensor, torch.Tensor, torch.Tensor]) Input flipped, instances flipped & BBs flipped
"""
# Flip input
input_flipped = input.flip(dims=(2, ))
# Flip instances
instances_flipped = instances.flip(dims=(2, ))
# Flip bounding boxes
image_center = torch.tensor((input.shape[2] // 2, input.shape[1] // 2))
bounding_boxes[:, [0, 2]] += 2 * (image_center -
bounding_boxes[:, [0, 2]])
bounding_boxes_w = torch.abs(bounding_boxes[:, 0] -
bounding_boxes[:, 2])
bounding_boxes[:, 0] -= bounding_boxes_w
bounding_boxes[:, 2] += bounding_boxes_w
return input_flipped, instances_flipped, bounding_boxes
def cummin_reversed(
tensor: torch.tensor, dim: int = 0) -> torch.tensor:
"""Return the cumulative minimum of the elements along a given axis in
inverse order.
Args:
tensor (torch.tensor): input tensor
dim (int, optional): Defaults to 0.
Axis along which the cumulative minimum is computed in inverse
order
Returns:
max_tensor (torch.tensor):
A tensor with the same size as the input holding the cumulative
minimum of the input tensor.
"""
return cummin(tensor.flip(dim), dim).flip(dim)
