def __init__(self, ksize: int):
"""
creates a pipeline that blurs the given frame using the median blur method
works very good for denoising purposes
:param ksize: the size of the kernel used by the filter, must be an odd number
:return: a pipeline that filters the image using the median blur method
"""
PipeLine.__init__(self, lambda frame: cv2.medianBlur(frame, ksize))
def __init__(self, ksize, iterations=1):
"""
creates a pipeline that erodes the image by a kernel of ones
used mainly for Erode & Dilate denoise filters
:param ksize: the kernel size, either an integer (meaning an nxn kernel) or a tuple (nxm kernel)
:param iterations: optional, the amount of Erode iterations to perform, default is 1
None! a large number of iterations will slow down the program
:return: a pipeline that erodes the given frame
"""
if type(ksize) is int:
ksize = (ksize, ksize)
PipeLine.__init__(
self, lambda frame: cv2.erode(
frame, np.ones(ksize), iterations=iterations))
def __init__(self, collision_func: Callable[[Shape, Shape], bool]):
def _filter(shapes: List[Shape]) -> List[Shape]:
filtered_shapes = []
for i, shape in enumerate(shapes):
shape_invalid = False
for j in range(i):
shape_invalid = collision_func(shape, shapes[j])
if shape_invalid:
break
if not shape_invalid:
filtered_shapes.append(shape)
return filtered_shapes
PipeLine.__init__(self, _filter)
def __init__(self, finding_func, color, drawing_func, *args, **kwargs):
def _draw(frame):
return drawing_func(frame, finding_func(frame), color, *args,
**kwargs)
PipeLine.__init__(self, _draw)
def __init__(self, ksize: Union[int, Tuple[int, int]], iterations=1):
PipeLine.__init__(self)
self.functions += (Erode(ksize, iterations) +
Dilate(ksize, iterations)).functions
def __init__(self, min_area: float):
PipeLine.__init__(self, lambda cnts: filter(lambda c: cv2.contourArea(c) >= min_area, cnts), list)
@PipeLine
def find_contours(frame):
return cv2.findContours(frame, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)[CONTOURS_INDEX]
@PipeLine
def sort_contours(cnts):
return sorted(cnts, key=lambda x: cv2.contourArea(x), reverse=True)
class FilterContours(PipeLine):
def __init__(self, min_area: float):
PipeLine.__init__(self, lambda cnts: filter(lambda c: cv2.contourArea(c) >= min_area, cnts), list)
convex_hull = PipeLine(cv2.convexHull)
@PipeLine
def convex_hull_multiple(cnts):
return list(map(convex_hull, cnts))
@PipeLine
def contour_center(cnt):
m = cv2.moments(cnt)
return int(m['m10'] / (m['m00'] + EPSILON)), int(m['m01'] / (m['m00'] + EPSILON))
@PipeLine
def contours_centers(cnts):
return list(map(contour_center, cnts))
def __init__(self, min_distance_ratio: float):
PipeLine.__init__(self)
from gbvision.models.basic_ops import normalized_distance_transform
self.functions += (normalized_distance_transform + ColorThreshold([[min_distance_ratio, 1.0]],
ColorThreshold.THRESH_TYPE_GRAY)).functions
from gbvision.utils.pipeline import PipeLine EMPTY_PIPELINE = PipeLine()
def __init__(self, ksize: int):
PipeLine.__init__(self, lambda frame: cv2.medianBlur(frame, ksize))
def __init__(self, ksize: Union[int, Tuple[int, int]], iterations=1):
if type(ksize) is int:
ksize = (ksize, ksize)
PipeLine.__init__(
self, lambda frame: cv2.dilate(
frame, np.ones(ksize), iterations=iterations))
def __mapper(func) -> PipeLine:
return PipeLine(lambda x: list(map(func, x)))
