def save_image(self, saved_path, name, inputs):
"""
save unet output as a form of image
"""
if not os.path.exists(saved_path):
os.makedirs(saved_path)
output = self.unet(inputs)
left = self.restore(inputs)
right = self.restore(output)
# The above two lines of code are wrong.To be precisely,errors will occur when the value of var left is less than
# the value of var right.For example,left=217,right=220,then result is 253 after abs operation.
diff = np.where(left > right, left - right,
right - left).clip(0, 255).astype(np.uint8)
plt.figure(num='unet result', figsize=(8, 8))
plt.subplot(2, 2, 1)
plt.title('source image')
plt.imshow(left)
plt.axis('off')
plt.subplot(2, 2, 2)
plt.title('unet output')
plt.imshow(right)
plt.axis('off')
plt.subplot(2, 2, 3)
plt.imshow(rgb2gray(diff), cmap='jet')
plt.colorbar(orientation='horizontal')
plt.title('difference in heatmap')
plt.axis('off')
plt.subplot(2, 2, 4)
plt.imshow(rgb2gray(diff.clip(0, 32)), cmap='jet')
plt.colorbar(orientation='horizontal')
plt.axis('off')
plt.tight_layout()
plt.savefig(add_prefix(saved_path, name))
plt.close()
def save_single_image(saved_path, model, name, inputs):
"""
save unet output as a form of image
"""
if not os.path.exists(saved_path):
os.makedirs(saved_path)
output = model(inputs)
left = restore(inputs)
right = restore(output)
diff = np.where(left > right, left - right,
right - left).clip(0, 255).astype(np.uint8)
plt.figure(num='unet result', figsize=(8, 8))
plt.subplot(2, 2, 1)
plt.title('source image')
plt.imshow(left)
plt.axis('off')
plt.subplot(2, 2, 2)
plt.title('unet output')
plt.imshow(right)
plt.axis('off')
plt.subplot(2, 2, 3)
plt.imshow(rgb2gray(diff), cmap='jet')
plt.colorbar(orientation='horizontal')
plt.title('difference in heatmap')
plt.axis('off')
plt.subplot(2, 2, 4)
plt.imshow(rgb2gray(diff.clip(0, 32)), cmap='jet')
plt.colorbar(orientation='horizontal')
plt.axis('off')
plt.tight_layout()
plt.savefig(add_prefix(saved_path, name))
plt.close()
def save_single_image(name, inputs, output, label, phase):
"""
save single image local
:param name: name
:param inputs: network input
:param output: network output
:param label: image label: 'lesion' or 'normal'
:param phase: image source: 'training' or 'validate' dataset
:return:
"""
left = restore(inputs)
right = restore(output)
plt.figure(num='unet result', figsize=(8, 8))
plt.subplot(2, 2, 1)
plt.title('source image')
plt.imshow(left)
plt.axis('off')
plt.subplot(2, 2, 2)
plt.title('output image')
plt.imshow(right)
plt.axis('off')
diff = np.where(left > right, left - right,
right - left).clip(0, 255).astype(np.uint8)
plt.subplot(2, 2, 3)
plt.imshow(rgb2gray(diff), cmap='jet')
plt.colorbar()
plt.tight_layout()
plt.savefig(add_prefix(args.prefix, '%s/%s/%s' % (phase, label, name)))
plt.close()
def save_single_image(label, name, inputs, output, unet, phase):
left = restore(inputs)
right = restore(unet)
# save network input image
plt.figure(num='unet result', figsize=(8, 8))
plt.subplot(2, 2, 1)
plt.title('source image: %s' % label)
plt.imshow(left)
plt.axis('off')
predict = F.softmax(output, dim=1)
predict = to_np(predict).flatten()
# save network output image
plt.subplot(2, 2, 2)
plt.title('lesion: %.2f, normal: %.2f' % (predict[0], predict[1]))
plt.imshow(right)
plt.axis('off')
# save difference directly
diff = np.where(left > right, left - right, right - left).clip(0, 255)
plt.subplot(2, 2, 3)
plt.imshow(rgb2gray(diff), cmap='jet')
plt.colorbar()
plt.title('difference in abs gray')
plt.tight_layout()
plt.savefig(add_prefix(args.prefix, '%s/%s/%s' % (phase, label, name)))
plt.close()
def contrast(self, saved_path, name, inputs):
"""
save unet output as a form of image
"""
if not os.path.exists(saved_path):
os.makedirs(saved_path)
if os.path.exists(add_prefix(saved_path, name)):
return
output = self.auto_encoder(inputs)
left = self.restore(inputs)
right = self.restore(output)
diff = np.where(left > right, left - right, right - left).clip(0, 255).astype(np.uint8)
plt.figure(num='unet result', figsize=(8, 8))
plt.subplot(2, 2, 1)
plt.title('source image')
plt.imshow(left)
plt.axis('off')
plt.subplot(2, 2, 2)
plt.title('unet output')
plt.imshow(right)
plt.axis('off')
plt.subplot(2, 2, 3)
plt.imshow(rgb2gray(diff), cmap='jet')
plt.colorbar(orientation='horizontal')
plt.title('difference in heatmap')
plt.axis('off')
plt.subplot(2, 2, 4)
plt.imshow(rgb2gray(diff.clip(0, 32)), cmap='jet')
plt.colorbar(orientation='horizontal')
plt.axis('off')
plt.tight_layout()
plt.savefig(add_prefix(saved_path, name))
print('file %s is saved to %s successfully.' %(name, add_prefix(saved_path, name)))
plt.close()
def calculate(self, image, name, binary_thresh):
left = evaluate.restore(image)
right = evaluate.restore(self.auto_encoder(image))
diff = np.where(left > right, left - right, right - left).clip(0, 255)
# binaryzation: background: 0 lesion areas: 1
_, binary = cv2.threshold(rgb2gray(diff).astype(np.uint8), binary_thresh, 1, cv2.THRESH_BINARY)
bounding_box_lst = self.groundtruth_dict[name]
dice_loss_lst = []
for bounding_box in bounding_box_lst:
pos_x, pos_y, size = bounding_box[0], bounding_box[1], bounding_box[2]
groundtruth = np.ones((size, size)).astype(np.uint8)
pred = binary[pos_y: pos_y + size, pos_x: pos_x + size]
dice_loss_lst.append(1 - distance.dice(groundtruth.reshape(-1), pred.reshape(-1)))
return sum(dice_loss_lst) / len(dice_loss_lst)
def screen(self):
return imresize(rgb2gray(self._screen)/255., self.dims)