def normalized_distance_to_center(filled_image): """regions outside of labeled image have normalized distance of 1""" distance_to_edge = distance_transform(np.pad(filled_image,1,'constant'))[1:-1,1:-1] max_distance = distance_to_edge.max() # median of all points furthest from edge center = tuple(np.median(np.where(distance_to_edge==max_distance),axis=1).astype(int)) mask = np.ones(filled_image.shape) mask[center[0],center[1]] = 0 distance_to_center = distance_transform(mask) return distance_to_center/(distance_to_center+distance_to_edge),center
GT_FMT = imset.lower()+'_%06d_pixeltruth.mat'
GT_MAT_NAME = 'truth_img'
res = .5
nb = int(50./res)
dist_hist = np.zeros((nb), dtype=np.uint64)
for idx in range(1, num_img+1):
logit_im = loadmat(join(PRED_PATH, LOGIT_FMT % idx))[LOGIT_MAT_NAME][...,1:].reshape(-1, num_classes)
gt = loadmat(join(GT_PATH, GT_FMT % idx))[GT_MAT_NAME]
fg = np.ones_like(gt)
fg[(gt==0) | (gt==255)] = 0
#weightmap = np.minimum(20, distance_transform(fg)).ravel()/20.
weightmap = distance_transform(fg).ravel()
for i, (tr, logits) in enumerate(zip(gt.ravel(), logit_im)):
if tr == bg_class or tr == 255: continue
pred = np.argmax(logits)
if tr-1 == pred: continue
dist_hist[min(int(np.floor(min(weightmap[i], 50)/res)), nb-1)] += 1
savemat('incorr_dist_hist.mat', {'dist_hist': dist_hist})
else:
dist_hist = loadmat('incorr_dist_hist.mat')['dist_hist']
import matplotlib.pyplot as plt
shape_features = {
'area' : lambda r: r.area,
'perimeter' : lambda r: r.perimeter,
'form_factor': lambda r:4*np.pi*r.area/(r.perimeter)**2, #isoperimetric quotient
'solidity': lambda r: r.solidity,
'extent': lambda r: r.extent,
'euler_number': lambda r: r.euler_number,
'centroid_r': lambda r: r.local_centroid[0], # ACTUALLY SHOULD BE POINT FARTHEST FROM EDGE?
'centroid_c': lambda r: r.local_centroid[1], # ACTUALLY SHOULD BE POINT FARTHEST FROM EDGE?
'eccentricity': lambda r: r.eccentricity,
'major_axis' : lambda r: r.major_axis_length,
'minor_axis' : lambda r: r.minor_axis_length,
'orientation' : lambda r: r.orientation,
'compactness' : lambda r: 2*np.pi*(r.moments_central[0,2]+r.moments_central[2,0])/(r.area**2),
'max_radius' : lambda r: distance_transform(np.pad(r.filled_image,1,'constant')).max(), #filled_image or image?
'median_radius' : lambda r: np.median(distance_transform(np.pad(r.filled_image,1,'constant'))[1:-1,1:-1][r.filled_image]), #filled_image or image?
'mean_radius' : lambda r: distance_transform(np.pad(r.filled_image,1,'constant'))[1:-1,1:-1][r.filled_image].mean(), #filled_image or image?
'feret_diameter' : lambda r: min_max_feret_diameter(r.coords),
# 'min_feret' : lambda r: minimum_feret_diameter(r.coords),
# 'max_feret' : lambda r: maximum_feret_diameter(r.coords),
'zernike' : lambda r: zernike_minimum_enclosing_circle(r.coords, degree=ZERNIKE_DEGREE) # cp/centrosome zernike divides zernike magnitudes by minimum enclosing circle magnitude; unclear why
}
zernike_nums = ['zernike_'+str(radial)+'_'+str(azimuthal)
for radial in range(ZERNIKE_DEGREE+1)
for azimuthal in range(radial%2,radial+2,2)]
shape_columns = {'zernike_'+str(num):zernike_num for num,zernike_num in enumerate(zernike_nums)}
shape_columns.update({
'feret_diameter_0':'min_feret_diameter',
def max_median_mean_radius(filled_image):
transformed = distance_transform(np.pad(filled_image, 1,
'constant'))[1:-1,
1:-1][filled_image]
return (transformed.max(), np.median(transformed), transformed.mean())