def compute_save_auc(data_set_name, image_pred_method, res_path, image_labels,
image_predictions, class_name):
'''
It computes and saves the results in a file. ROC curve and confusion matrix visualizations are also done here.
:param data_set_name:
:param image_pred_method:
:param res_path:
:param image_labels:
:param image_predictions:
:param class_name:
:return:
'''
auc_all_classes_v1, fpr, tpr, roc_auc = compute_auc_1class(
image_labels, image_predictions)
save_evaluation_results([class_name], auc_all_classes_v1, 'auc_prob_' +
data_set_name + '_' + image_pred_method + '.csv',
res_path)
plot_roc_curve(fpr, tpr, roc_auc, data_set_name, res_path)
conf_matrix = confusion_matrix(
image_labels, np.array(image_predictions > 0.5, dtype=np.float32))
plot_confusion_matrix(conf_matrix, [0, 1],
res_path,
data_set_name,
normalize=False,
title=None)
plot_confusion_matrix(conf_matrix, [0, 1],
res_path,
data_set_name + 'norm',
normalize=True,
title=None)
def compute_save_inst_auc_results(data_set_name, res_path, inst_auc):
mean_auc = np.mean(inst_auc, axis=0)
print("Instance AUC")
print(mean_auc)
save_evaluation_results(["inst_auc"],
mean_auc,
"inst_auc_" + data_set_name + '.csv',
res_path,
add_col=None,
add_value=None)
def compute_save_dice_results(data_set_name, res_path, has_bbox, dice_scores):
dice_score_ma = np.ma.masked_array(dice_scores,
mask=np.equal(dice_scores, -1))
mean_dice = np.mean(dice_score_ma, axis=0)
print("DICE")
print(mean_dice)
save_evaluation_results(["dice"],
mean_dice,
"dice_" + data_set_name + '.csv',
res_path,
add_col=None,
add_value=None)
def compute_save_accuracy_results(data_set_name, res_path, has_bbox,
acc_localization):
print("accuracy bbox present vs accurate")
total_accurate_segmentations = np.sum(acc_localization, axis=0)
total_segmentation = np.sum(has_bbox, axis=0)
with np.errstate(divide='ignore', invalid='ignore'):
acc_class = total_accurate_segmentations / total_segmentation
print("ACCURACY RESULTS FROM BBOX")
print(acc_class)
save_evaluation_results(["accuracy"],
acc_class,
"accuracy_" + data_set_name + '.csv',
res_path,
add_col=None,
add_value=None)
def compute_save_dice_results(eval_df, data_set_name, res_path, dice_scores):
dice_score_ma = np.ma.masked_array(dice_scores, mask=np.equal(dice_scores, -1))
mean_dice = np.mean(dice_score_ma, axis=0)
print("DICE")
print(mean_dice)
# save_evaluation_results(["dice"], mean_dice, "dice_" + data_set_name + '.csv', res_path,
# add_col=None, add_value=None)
return save_evaluation_results(eval_df, ["dice"], mean_dice, "evaluation_performance_" + data_set_name + '.csv',
res_path,
add_col=None, add_value=None)