from .util import dtype_type_check,\
interpolation_type_check,\
channel_type_check,\
get_cv2_interp_type
from .imports import import_opencv
from .blocks import ImageBlock
cv2 = import_opencv()
import numpy as np
import imagepypelines as ip
[docs]class Resize(ImageBlock):
"""splits images into separate component channels
Attributes:
w_scale_type(str): type of scaling used for image width, either
"proportional" or "absolute"
h_scale_type(str): type of scaling used for image height, either
"proportional" or "absolute"
h_param(int,float): vertical scale or absolute height to resize
image to
w_param(int,float): horizontal scale or absolute height to resize
image to
interp(str): interpolation type for resizing. One of
'nearest', 'linear', 'area', 'cubic', 'lanczos4'
Default Enforcement:
1) image
type: np.ndarray
shapes: [(None,None,None),(None,None)]
notes: image must be ordered [height,width,channels]
Batch Size:
"each"
"""
def __init__(self, h=None, w=None, scale_h=None, scale_w=None, interp='nearest'):
"""Instantiates the object
Args:
w(None,int): width to scale image to, must be None is scale_w is
defined
h(None,int): height to scale image to, must be None is scale_h is
defined
scale_h(None,float): vertical scale for the image, must be None
if 'h' is defined
scale_w(None,float): horizontal scale for the image, must be None
if 'w' is defined
interp(str): interpolation type for image scaling, must be one of:
'nearest', 'linear', 'area', 'cubic', 'lanczos4'
"""
super().__init__(order="HWC")
# make sure either h or scale_h is defined
if (h is None) and (scale_h is None):
raise ValueError("'h' or 'scale_h' must be defined")
# make sure either w or scale_w is defined
if (w is None) and (scale_w is None):
raise ValueError("'w' or 'scale_w' must be defined")
# make sure only h or scale_h is defined
if (not h is None) and (not scale_h is None):
raise ValueError("only 'h' or 'scale_h' can be defined")
# make sure only w or scale_w is defined
if (not w is None) and (not scale_w is None):
raise ValueError("only 'w' or 'scale_w' can be defined")
# set w instance variables
if w is None:
self.w_scale_type = 'proportional'
self.w_param = scale_w
else:
self.w_scale_type = 'absolute'
self.w_param = w
# set h instance variables
if h is None:
self.h_scale_type = 'proportional'
self.h_param = scale_h
else:
self.h_scale_type = 'absolute'
self.h_param = h
self.__cv2_interp = get_cv2_interp_type(interp)
self.interp = interp
self.enforce('image', np.ndarray, [(None,None,None),(None,None)])
[docs] def process(self, image):
"""Resizes the image to the specified dimensions
Args:
image(np.ndarray): image to resize, must be shaped
[height,width,channels]
Returns:
np.ndarray: resized image
"""
# get h dimension
if self.h_scale_type == "proportional":
new_h = round(self.h_param * image.shape[0], 0)
else:
new_h = self.h_param
# get w dimension
if self.w_scale_type == "proportional":
new_w = round(self.w_param * image.shape[1], 0)
else:
new_w = self.w_param
return cv2.rezize(image, (new_w,new_h), interpolation=self.__cv2_interp)
# END
