def get_annotation_suggestions(self, n_predicts=0, n_samples=100):
n_digits = len(repr(abs(n_samples)))
predictions, save_names = self.predict(n_predicts=n_predicts)
images = [[ImageHandler._tensor_to_image(img, mask=True) for img in imgs] for imgs in predictions]
# sum_images = np.sum(images, axis=0)
result_imgs = []
for imgs in images:
result_and = np.zeros_like(imgs[0])
result_or = np.zeros_like(imgs[0])
for img in imgs:
result_and = np.logical_and(img, result_and)
result_or = np.logical_or(img, result_or)
result_imgs.append(result_or - result_and)
uncertainties = []
for idx, res_img in result_imgs:
uncertainties.append((np.sum(res_img), idx))
uncertainties.sort(key=lambda tup: tup[0], reverse=True)
if n_samples < len(uncertainties):
n_samples = len(uncertainties)
output_dir = self.options.output_path + "/" + self.options.name + "/Suggest/" + "Epoch_" + str(self.options.load_epoch)
for i in range(n_samples):
self.image_handler.save_image(result_imgs[uncertainties[i][1]], output_dir, save_names[uncertainties[i][1]])
