Table 1 Test results on all instances.

From: WormSwin: Instance segmentation of C. elegans using vision transformer

 

\({\hbox {AP}_{0.50}}\)

\({\hbox {AP}_{0.75}}\)

\({\hbox {mAP}_{0.50:0.95}}\)

CSB-1

 WormSwin (box)

0.990

0.976

0.819

 WormSwin (mask)

0.990

0.675

0.585

Synthetic

 WormSwin (box)

0.989

0.978

0.909

 WormSwin (mask)

0.977

0.918

0.679

BBBC010

 PatchPerPix ppp+dec (mask)

0.939

0.891

0.775

 WormSwin (box)†

0.985

0.949

0.823

 WormSwin (mask)†

0.954

0.801

0.622

 WormSwin (mask)*, †

0.964

0.815

0.629

MD

 WormSwin (box)†

0.990

0.968

0.832

 WormSwin (mask)†

0.980

0.551

0.542

  1. “Box” and “mask” refer to the accuracy of detection of the bounding box and segmentation mask, respectively. PatchPerPix ppp+dec refers to the network variant, introduced by19. (*Multi-scale testing, †additional training data).
Back to article page