Evaluation of semantic segmentation

Parameters to evaluate the semantic segmentation

• Execution time

• Memory footprint

• Accuracy

• Trained model hardware

pixel accuracy

mean pixel accuracy

IOU/jaccard_score

import numpy as np from sklearn.metrics import jaccard_score

y_pred = [0, 5, 1, 2, 4] y_true = [0, 5, 2, 2, 2] print(jaccard_score(y_true, y_pred, average=None)) print(jaccard_score(y_true, y_pred, average='micro')) print(jaccard_score(y_true, y_pred, average='macro')) [1. 0. 0.33333333 0. 1. ] shape = [n_unique_labels] 0.42857142857142855 0.4666666666666666

Dice coefficient

import numpy as np

k=1

#segmentation seg = np.zeros((100,100), dtype='int') seg[30:70, 30:70] = k

#ground truth gt = np.zeros((100,100), dtype='int') gt[30:70, 40:80] = k

dice = np.sum(seg[gt==k])*2.0 / (np.sum(seg) + np.sum(gt))

print 'Dice similarity score is {}'.format(dice)