Table 4 Evaluation of different fusion strategies for combining dual-stream features. Progressive fusion with attention shows optimal performance.

From: Bilateral collaborative streams with multi-modal attention network for accurate polyp segmentation

Fusion strategy

Endoscene

ClinicDB

ColonDB

ETIS

Kvasir-SEG

Avg.

Simple concatenation

0.863

0.912

0.771

0.742

0.897

0.837

Element-wise addition

0.869

0.918

0.778

0.748

0.903

0.843

Feature pyramid network

0.881

0.925

0.789

0.761

0.911

0.853

Progressive fusion

Without attention

0.887

0.931

0.796

0.773

0.916

0.861

With PAB only

0.902

0.936

0.807

0.785

0.923

0.871

With CBAM only

0.895

0.933

0.801

0.778

0.920

0.865

With PAB + CBAM

0.908

0.941

0.821

0.810

0.929

0.882

Back to article page