def test_smoke(self, device, dtype):
# Sample two points of the rectangle
points = torch.rand(1, 4, device=device, dtype=dtype)
# Fill acording missing points
bbox = torch.zeros(1, 4, 2, device=device, dtype=dtype)
bbox[0, 0] = points[0][:2]
bbox[0, 1, 0] = points[0][2]
bbox[0, 1, 1] = points[0][1]
bbox[0, 2] = points[0][2:]
bbox[0, 3, 0] = points[0][0]
bbox[0, 3, 1] = points[0][3]
# Validate
assert validate_bbox(bbox)
def crop_by_boxes(
tensor: torch.Tensor,
src_box: torch.Tensor,
dst_box: torch.Tensor,
mode: str = 'bilinear',
padding_mode: str = 'zeros',
align_corners: bool = True,
validate_boxes: bool = True
) -> torch.Tensor:
"""Perform crop transform on 2D images (4D tensor) given two bounding boxes.
Given an input tensor, this function selected the interested areas by the provided bounding boxes (src_box).
Then the selected areas would be fitted into the targeted bounding boxes (dst_box) by a perspective transformation.
So far, the ragged tensor is not supported by PyTorch right now. This function hereby requires the bounding boxes
in a batch must be rectangles with same width and height.
Args:
tensor: the 2D image tensor with shape (B, C, H, W).
src_box: a tensor with shape (B, 4, 2) containing the coordinates of the bounding boxes
to be extracted. The tensor must have the shape of Bx4x2, where each box is defined in the clockwise
order: top-left, top-right, bottom-right and bottom-left. The coordinates must be in x, y order.
dst_box: a tensor with shape (B, 4, 2) containing the coordinates of the bounding boxes
to be placed. The tensor must have the shape of Bx4x2, where each box is defined in the clockwise
order: top-left, top-right, bottom-right and bottom-left. The coordinates must be in x, y order.
mode: interpolation mode to calculate output values
``'bilinear'`` | ``'nearest'``.
padding_mode: padding mode for outside grid values
``'zeros'`` | ``'border'`` | ``'reflection'``.
align_corners: mode for grid_generation.
validate_boxes: flag to perform validation on boxes.
Returns:
torch.Tensor: the output tensor with patches.
Examples:
>>> input = torch.arange(16, dtype=torch.float32).reshape((1, 1, 4, 4))
>>> src_box = torch.tensor([[
... [1., 1.],
... [2., 1.],
... [2., 2.],
... [1., 2.],
... ]]) # 1x4x2
>>> dst_box = torch.tensor([[
... [0., 0.],
... [1., 0.],
... [1., 1.],
... [0., 1.],
... ]]) # 1x4x2
>>> crop_by_boxes(input, src_box, dst_box, align_corners=True)
tensor([[[[ 5.0000, 6.0000],
[ 9.0000, 10.0000]]]])
Note:
If the src_box is smaller than dst_box, the following error will be thrown.
RuntimeError: solve_cpu: For batch 0: U(2,2) is zero, singular U.
"""
if validate_boxes:
validate_bbox(src_box)
validate_bbox(dst_box)
if len(tensor.shape) != 4:
raise AssertionError(f"Only tensor with shape (B, C, H, W) supported. Got {tensor.shape}.")
# compute transformation between points and warp
# Note: Tensor.dtype must be float. "solve_cpu" not implemented for 'Long'
dst_trans_src: torch.Tensor = get_perspective_transform(src_box.to(tensor), dst_box.to(tensor))
bbox: Tuple[torch.Tensor, torch.Tensor] = infer_bbox_shape(dst_box)
if not ((bbox[0] == bbox[0][0]).all() and (bbox[1] == bbox[1][0]).all()):
raise AssertionError(
f"Cropping height, width and depth must be exact same in a batch. "
f"Got height {bbox[0]} and width {bbox[1]}."
)
h_out: int = int(bbox[0][0].item())
w_out: int = int(bbox[1][0].item())
return crop_by_transform_mat(
tensor, dst_trans_src, (h_out, w_out), mode=mode, padding_mode=padding_mode, align_corners=align_corners
)
def _boxes_to_quadrilaterals(boxes: torch.Tensor,
mode: str = "xyxy",
validate_boxes: bool = True) -> torch.Tensor:
"""Convert from boxes to quadrilaterals."""
mode = mode.lower()
if mode.startswith("vertices"):
batched = boxes.ndim == 4
if not (3 <= boxes.ndim <= 4
and boxes.shape[-2:] == torch.Size([4, 2])):
raise ValueError(
f"Boxes shape must be (N, 4, 2) or (B, N, 4, 2) when {mode} mode. Got {boxes.shape}."
)
elif mode.startswith("xy"):
batched = boxes.ndim == 3
if not (2 <= boxes.ndim <= 3 and boxes.shape[-1] == 4):
raise ValueError(
f"Boxes shape must be (N, 4) or (B, N, 4) when {mode} mode. Got {boxes.shape}."
)
else:
raise ValueError(f"Unknown mode {mode}")
boxes = boxes if boxes.is_floating_point() else boxes.float()
boxes = boxes if batched else boxes.unsqueeze(0)
if mode.startswith("vertices"):
if mode == "vertices":
# Avoid passing reference
quadrilaterals = boxes.clone()
elif mode == "vertices_plus":
quadrilaterals = boxes.clone() # TODO: perform +1
else:
raise ValueError(f"Unknown mode {mode}")
validate_boxes or validate_bbox(quadrilaterals)
elif mode.startswith("xy"):
if mode == "xyxy":
height, width = boxes[..., 3] - boxes[...,
1], boxes[...,
2] - boxes[..., 0]
elif mode == "xyxy_plus":
height, width = boxes[..., 3] - boxes[..., 1] + 1, boxes[
..., 2] - boxes[..., 0] + 1
elif mode == "xywh":
height, width = boxes[..., 3], boxes[..., 2]
else:
raise ValueError(f"Unknown mode {mode}")
if validate_boxes:
if (width <= 0).any():
raise ValueError("Some boxes have negative widths or 0.")
if (height <= 0).any():
raise ValueError("Some boxes have negative heights or 0.")
xmin, ymin = boxes[..., 0], boxes[..., 1]
quadrilaterals = _boxes_to_polygons(xmin, ymin, width, height)
else:
raise ValueError(f"Unknown mode {mode}")
quadrilaterals = quadrilaterals if batched else quadrilaterals.squeeze(0)
return quadrilaterals