def test_imfuse():
"""Test function for imfuse()
"""
# path and name of test files
im1_name = os.path.join(data_path, "left.png")
im2_name = os.path.join(data_path, "right.png")
# read test images and their masks
im1 = cv2.imread(im1_name)
im2 = cv2.imread(im2_name)
# fuse images
imf = pymg.imfuse(im1, im2)
# plot images
plt.close('all')
plt.subplot(221)
plt.imshow(im1)
plt.title("Left image")
plt.subplot(222)
plt.imshow(im2)
plt.title("Right image")
plt.subplot(212)
plt.imshow(imf)
plt.title("imfuse")
plt.show(block=False)
preddice_test_pred = dice_model.predict(
im_test[i, :, :, :].reshape((1, ) + im_test.shape[1:]))
# plot results
plt.clf()
plt.subplot(321)
plt.imshow(im_test[i, :, :, :])
plt.title('histology, i = ' + str(i))
plt.subplot(323)
plt.imshow(seg_test[i, :, :, 0])
plt.title('ground truth contours')
plt.subplot(324)
aux = cv2.dilate(predseg_test[i, :, :, 0],
kernel=np.ones(shape=(3,
3))) # dilate for better visualisation
plt.imshow(pystoim.imfuse(seg_test[i, :, :, 0], aux).astype(np.float32))
plt.title('ground truth (gree) vs. \npredicted (purple) contours')
plt.subplot(325)
plt.imshow(preddice_test[i, :, :, 0], cmap='Greys_r')
plt.title('ground truth Dice coeff')
plt.subplot(326)
plt.imshow(preddice_test_pred[0, :, :, 0])
plt.title('estimated Dice coeff')
# visualise results
i = 18
# run image through network
preddice_test_pred = dice_model.predict(
im_test[i, :, :, :].reshape((1, ) + im_test.shape[1:]))
qual = cytometer.utils.match_overlapping_labels(
labels_test=labels[i, :, :, 0],
labels_ref=reflab[i, :, :, 0])
labels_qual_augmented = cytometer.utils.paint_labels(
labels=labels_augmented,
paint_labs=qual['lab_test'],
paint_values=qual['dice'])
# compare randomly transformed histology to corresponding Dice coefficient
plt.clf()
plt.subplot(321)
plt.imshow(aux_dataset['im'][0, :, :, :])
plt.subplot(322)
plt.imshow(labels_qual_augmented, cmap='Greys_r')
plt.subplot(323)
aux = pystoim.imfuse(aux_dataset['im'][0, :, :, :],
labels_qual_augmented)
plt.imshow(aux, cmap='Greys_r')
plt.subplot(324)
plt.imshow(labels_augmented)
plt.subplot(325)
plt.imshow(labels_borders_augmented)
# filenames for the Dice coefficient augmented files
predlab_file = os.path.join(
training_augmented_dir,
base_name.replace(
'im_seed_nan_', 'predlab_kfold_' + str(fold_i).zfill(2) +
'_seed_' + str(seed).zfill(3) + '_'))
predseg_file = os.path.join(
training_augmented_dir,
base_name.replace(
mask = mask.astype(np.float32)
if DEBUG:
for i in range(n_im):
print(' ** Image: ' + str(i) + '/' + str(n_im - 1))
plt.clf()
plt.subplot(221)
plt.imshow(im[i, :, :, :])
plt.subplot(222)
plt.imshow(dmap[i, :, :, 0])
plt.subplot(223)
plt.imshow(mask[i, :, :, 0])
plt.subplot(224)
a = im[i, :, :, :]
b = mask[i, :, :, 0]
plt.imshow(pystoim.imfuse(a, b))
plt.show()
'''Receptive field
'''
# list of model files to inspect
model_files = glob.glob(os.path.join(saved_models_dir, model_name))
receptive_field_size = []
for model_file in model_files:
print(model_file)
# estimate receptive field of the model
def model_build_func(input_shape):