Spaces:
Runtime error
Runtime error
File size: 5,172 Bytes
0ce2b9f | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 | # should work for all apporaches (Deep learning or not)
# should work on segmentation masks
# should include a smoothing and an max-area threshold
# should try to work with the tortina circle (or maybe better estimate it again)
# maybe even to apply before thresholding, directly on anomaly maps/other outputs to have better smoothing
# output should be again a segmentation mask
import numpy as np
import cv2
def get_points_in_circle(circle):
x0, y0, radius = circle.astype(np.int32)
x_ = np.arange(x0 - radius - 1, x0 + radius + 1, dtype=np.int32)
y_ = np.arange(y0 - radius - 1, y0 + radius + 1, dtype=np.int32)
x, y = np.where((x_[:,np.newaxis] - x0)**2 + (y_ - y0)**2 <= radius**2)
# for x, y in zip(x_[x], y_[y]): # yield x, y
return (x_[x], y_[y])
def find_single_tortina_circle(image):
height, width, num_channels = image.shape
#find the biggest circle (tortina) in the image
#init with default circle
selected_circle = np.array((width//2, height//2, int(height/2*0.9)))
try:
#circles = find_tortinas(image, 1, force_num_tortinas=True)
height, width = image.shape[:2]
if (image.ndim == 2) or (image.shape[-1] == 1):
gray = image
else:
gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
blur = cv2.blur(gray, (7,7))
circles = cv2.HoughCircles(blur, cv2.HOUGH_GRADIENT, 3, minDist=width//2)
if circles is None:
circles = []
else:
circles = circles[0].astype(np.uint32)
except:
circles = []
if len(circles) > 0:
max_r = 0
for x,y,r in circles:
if r > max_r:
max_r = r
selected_circle = np.array((x,y,r))
return selected_circle
def postprocessing(image, segmask, fat_bloom_id=1):
if segmask.shape != image.shape[:2]:
raise ValueError(
"""segmask argument should represent a segmentation mask with 2 dimensions (height, width)!
This means that its values should already be thresholded and (in case of rgb) reduced to
a single channel.
""")
height, width = image.shape[:2]
selected_circle = find_single_tortina_circle(image)
#debugging
#plot_image_with_circle(image, selected_circle)
# smooth the segmask
binary = (segmask == fat_bloom_id).astype(np.uint8)
kernel = np.ones((5,5),np.uint8)
closing = cv2.morphologyEx(binary, cv2.MORPH_CLOSE, kernel)
opening = cv2.morphologyEx(closing, cv2.MORPH_OPEN, kernel)
#the opening values should overwrite the fat_bloom of the input
#segmask. Where we had fat bloom in the input but not anymore in the
#opening we need to determine whether it is backgorund or tortina. We
#can do so using the estimated circle.
#initialize everything to background. background = 0
new_segmask = np.zeros_like(segmask)
tortina_indices = get_points_in_circle(selected_circle)
#correction
out_of_width_0 = (tortina_indices[0] < 0)
out_of_width_1 = (tortina_indices[0] >= width)
out_of_height_0 = (tortina_indices[1] < 0)
out_of_height_1 = (tortina_indices[1] >= height)
tortina_indices[0][out_of_width_0] = 0
tortina_indices[0][out_of_width_1] = width - 1
tortina_indices[1][out_of_height_0] = 0
tortina_indices[1][out_of_height_1] = height - 1
# remove_indices = (tortina_indices[0] < 0) | (tortina_indices[0] >= height) | \
# (tortina_indices[1] < 0) | (tortina_indices[1] >= width)
# keep_indices = np.setdiff1d(np.arange(len(tortina_indices[0])), remove_indices)
# tortina_indices = (tortina_indices[0][keep_indices], tortina_indices[1][keep_indices])
is_tortina = np.zeros_like(segmask, dtype=bool)
is_tortina[tortina_indices] = True
#tortina = 1
new_segmask[is_tortina] = 1
#fat_bloom = 2
new_segmask[(opening == 1) & is_tortina] = 2
return new_segmask
def final_prediction(anomaly_map, segmask):
fat_bloom_area = np.count_nonzero(segmask == 2)
tortina_area = np.count_nonzero(segmask == 1) + fat_bloom_area
relative_area = fat_bloom_area / tortina_area
area = 0
if (relative_area > 0) and (relative_area <= 0.25):
area = 1
elif (relative_area > 0.25) and (relative_area <= 0.5):
area = 2
elif (relative_area > 0.5) and (relative_area <= 0.75):
area = 3
elif relative_area > 0.75:
area = 4
relative_intensity = 0.0
if fat_bloom_area > 0.0:
relative_intensity = anomaly_map[segmask==2].mean()
#TODO: intensity should be depending on colour of underlying tortina
# i.e. for a darker tortina intesity is automatically higher,
# but that should be relativiert.
intensity = 0
if (relative_intensity > 0) and (relative_intensity <= 0.25):
intensity = 1
elif (relative_intensity > 0.25) and (relative_intensity <= 0.5):
intensity = 2
elif (relative_intensity > 0.5) and (relative_intensity <= 0.75):
intensity = 3
elif relative_intensity > 0.75:
intensity = 4
return area, intensity |