def resize_shape_func(attrs, inputs, _):
"""
Shape function for dyn.image.resize op.
"""
layout = attrs.layout
if nchw_pack_layout(layout) or nchw_xc_layout(layout):
out = [
_resize_shape_func(inputs[0].shape, inputs[1],
convert(len(inputs[0].shape)), convert(2),
convert(3))
]
else:
height_axis = width_axis = 1
for i, letter in enumerate(layout):
if letter == "H":
height_axis = i
if letter == "W":
width_axis = i
out = [
_resize_shape_func(
inputs[0].shape,
inputs[1],
convert(len(inputs[0].shape)),
convert(height_axis),
convert(width_axis),
)
]
return out
def get_2d_pixel(data, layout, image_height, image_width, n, c, y, x, cc, ib, ic):
"""Get 2d pixel"""
y = tvm.te.max(tvm.te.min(y, image_height - 1), 0)
x = tvm.te.max(tvm.te.min(x, image_width - 1), 0)
if layout == "NHWC":
return data(n, y, x, c).astype("float")
if layout == "NCHW":
return data(n, c, y, x).astype("float")
if nchw_pack_layout(layout):
return data(n, c, y, x, ib, ic).astype("float")
# else must be NCHWxc
assert nchw_xc_layout(layout)
return data(n, c, y, x, cc).astype("float")
def get_2d_indices(indices, layout="NCHW"):
"""Get 2d indices"""
(cc, inum, ic) = (0, 0, 0)
if layout == "NHWC":
n, y, x, c = indices
cc = None
elif layout == "NCHW":
n, c, y, x = indices
cc = None
elif nchw_pack_layout(layout):
n, c, y, x, inum, ic = indices
else:
# else must be NCHWxc
assert nchw_xc_layout(layout)
n, c, y, x, cc = indices
return n, c, y, x, cc, inum, ic
def resize2d(
data,
roi,
size,
layout="NCHW",
method="linear",
coordinate_transformation_mode="half_pixel",
rounding_method="",
bicubic_alpha=-0.5,
bicubic_exclude=0,
extrapolation_value=0.0,
out_dtype=None,
output_shape=None,
):
"""Perform resize operation on the data.
Parameters
----------
data : tvm.te.Tensor
inputs is a 4-D tensor with shape
[batch, channel, in_height, in_width]
or [batch, in_height, in_width, channel]
roi: Tuple of Float or Expr
The region of interest for cropping the input image. Expected to be of
size 4, and format [start_h, start_w, end_h, end_w].
Only used if coordinate_transformation_mode is tf_crop_and_resize.
size: Tuple
Output resolution scale to
layout: string, optional
"NCHW", "NHWC", or "NCHWc".
coordinate_transformation_mode: string, optional
Describes how to transform the coordinate in the resized tensor
to the coordinate in the original tensor.
Refer to the ONNX Resize operator specification for details.
Available options are "half_pixel", "align_corners" and "asymmetric".
method: string, optional
method of interpolation ("nearest", "linear", "bicubic")
coordinate_transformation_mode : string, optional
Describes how to transform the coordinate in the resized tensor
to the coordinate in the original tensor.
[half_pixel, align_corners, asymmetric, pytorch_half_pixel,
tf_half_pixel_for_nn, and tf_crop_and_resize].
rounding_method:
Method for rounding coordinate locations
bicubic_alpha: float, optional
Bicubic spline coefficient
bicubic_exclude: bool, optional:
Exclude values outside the image fdor bicubic interpolation
extrapolation_value: float, optional
Value used for extrapolation, when applicable.
out_dtype: string, optional
Type to return. If left None will be same as input type.
output_shape: tvm.tir.container.Array, optional
Shape to return. If left None will be inferred
(If shape is determined dynamically, pass out_dtype.shape as output_shape)
Returns
-------
output : tvm.te.Tensor
4-D with shape [batch, channel, in_height*scale, in_width*scale]
or [batch, in_height*scale, in_width*scale, channel]
or 5-D with shape [batch, channel-major, in_height*scale, in_width*scale, channel-minor]
"""
method = method.lower()
if layout == "NHWC":
in_n, in_h, in_w, in_c = data.shape
if output_shape is None:
output_shape = [in_n, size[0], size[1], in_c]
elif layout == "NCHW":
in_n, in_c, in_h, in_w = data.shape
if output_shape is None:
output_shape = [in_n, in_c, size[0], size[1]]
elif nchw_pack_layout(layout): # for NCHWinic
in_n, in_c, in_h, in_w, in_inum, in_ic = data.shape
if output_shape is None:
output_shape = [in_n, in_c, size[0], size[1], in_inum, in_ic]
elif nchw_xc_layout(layout): # for NCHWxc
in_n, in_c, in_h, in_w, in_cc = data.shape
if output_shape is None:
output_shape = [in_n, in_c, size[0], size[1], in_cc]
else:
raise ValueError("%s layout is not supported." % layout)
if isinstance(size, tuple):
size = list(size)
for i in range(2):
if isinstance(size[i], int):
size[i] = tvm.tir.IntImm("int32", size[i])
def compute_func(*indices):
return _resize_2d(
indices,
data,
roi,
in_h,
in_w,
size[0],
size[1],
method=method,
layout=layout,
coordinate_transformation_mode=coordinate_transformation_mode,
rounding_method=rounding_method,
alpha=bicubic_alpha,
exclude_outside=bicubic_exclude,
extrapolation_value=extrapolation_value,
out_dtype=out_dtype,
)
return te.compute(output_shape,
compute_func,
name="resize",
tag=tag.INJECTIVE)
def resize(
data,
size,
layout="NCHW",
method="bilinear",
coordinate_transformation_mode="half_pixel",
rounding_method="",
bicubic_alpha=-0.5,
bicubic_exclude=0,
out_dtype=None,
output_shape=None,
):
"""Perform resize operation on the data.
Parameters
----------
data : tvm.te.Tensor
inputs is a 4-D tensor with shape
[batch, channel, in_height, in_width]
or [batch, in_height, in_width, channel]
size: Tuple
Output resolution scale to
layout: string, optional
"NCHW", "NHWC", or "NCHWc".
coordinate_transformation_mode: string, optional
Describes how to transform the coordinate in the resized tensor
to the coordinate in the original tensor.
Refer to the ONNX Resize operator specification for details.
Available options are "half_pixel", "align_corners" and "asymmetric".
method: {"bilinear", "nearest_neighbor", "bicubic"}
Method to be used for resizing.
out_dtype: string, optional
Type to return. If left None will be same as input type.
output_shape: tvm.tir.container.Array, optional
Shape to return. If left None will be inferred
(If shape is determined dynamically, pass out_dtype.shape as output_shape)
Returns
-------
output : tvm.te.Tensor
4-D with shape [batch, channel, in_height*scale, in_width*scale]
or [batch, in_height*scale, in_width*scale, channel]
or 5-D with shape [batch, channel-major, in_height*scale, in_width*scale, channel-minor]
"""
method = method.lower()
if layout == "NHWC":
in_n, in_h, in_w, in_c = data.shape
if output_shape is None:
output_shape = [in_n, size[0], size[1], in_c]
elif layout == "NCHW":
in_n, in_c, in_h, in_w = data.shape
if output_shape is None:
output_shape = [in_n, in_c, size[0], size[1]]
elif nchw_pack_layout(layout): # for NCHWinic
in_n, in_c, in_h, in_w, in_inum, in_ic = data.shape
if output_shape is None:
output_shape = [in_n, in_c, size[0], size[1], in_inum, in_ic]
elif nchw_xc_layout(layout): # for NCHWxc
in_n, in_c, in_h, in_w, in_cc = data.shape
if output_shape is None:
output_shape = [in_n, in_c, size[0], size[1], in_cc]
else:
raise ValueError("%s layout is not supported." % layout)
if isinstance(size, tuple):
size = list(size)
for i in range(2):
if isinstance(size[i], int):
size[i] = tvm.tir.IntImm("int32", size[i])
def _nearest_neighbor(*indices):
return resize_nearest_neighbor(
indices,
data,
in_h,
in_w,
size[0],
size[1],
layout=layout,
coordinate_transformation_mode=coordinate_transformation_mode,
rounding_method=rounding_method,
out_dtype=out_dtype,
)
def _bilinear(*indices):
return resize_bilinear(
indices,
data,
in_h,
in_w,
size[0],
size[1],
layout=layout,
coordinate_transformation_mode=coordinate_transformation_mode,
out_dtype=out_dtype,
)
def _bicubic(*indices):
return resize_bicubic(
indices,
data,
in_h,
in_w,
size[0],
size[1],
layout=layout,
coordinate_transformation_mode=coordinate_transformation_mode,
out_dtype=out_dtype,
alpha=bicubic_alpha,
exclude_outside=bicubic_exclude,
)
# Determine which interpolation method to use then run it.
if method == "nearest_neighbor":
compute_func = _nearest_neighbor
elif method == "bilinear":
compute_func = _bilinear
elif method == "bicubic":
compute_func = _bicubic
else:
raise ValueError("%s method is not supported." % method)
return te.compute(output_shape,
compute_func,
name="resize",
tag=tag.INJECTIVE)
