Skip to main content

Table 2 Evaluation on the 5-class and 10-class standard and hard test sets of SegSemDTD

From: Semantic segmentation of textured mosaics

Metric

Class

Experiment

  

5-class

10-class

5-class (hard)

10-class (hard)

PA

All

0.9945

0.9924

0.9622

0.8228

mPA

All

0.9945

0.9923

0.9621

0.8244

mIoU

All

0.9891

0.9849

0.9281

0.7088

\(\text {IoU}_1\)

Banded

0.9907

0.9896

0.9450

0.6640

\(\text {IoU}_2\)

Chequered

0.9878

0.9872

0.9654

0.8735

\(\text {IoU}_3\)

Woven

0.9903

0.9908

0.9004

0.5831

\(\text {IoU}_4\)

Wrinkled

0.9883

0.9884

0.8713

0.7100

\(\text {IoU}_5\)

Fibrous

0.9884

0.9832

0.9583

0.7064

\(\text {IoU}_6\)

Cracked

N/A

0.9872

N/A

0.7827

\(\text {IoU}_7\)

Grid

N/A

0.9886

N/A

0.6275

\(\text {IoU}_8\)

Honeycombed

N/A

0.9761

N/A

0.4594

\(\text {IoU}_9\)

Polka-dotted

N/A

0.9852

N/A

0.9350

\(\text {IoU}_{10}\)

Zigzagged

N/A

0.9732

N/A

0.7462

  1. Best and worst \(\text {IoU}_i\) shown in bold and italics, respectively