Table 2 Comparison of computational complexity and performance on MIRFlickr.

From: Enhancing cross-modal retrieval via label graph optimization and hybrid loss functions

Method

Params (M)

FLOPs (G)

Training time

mAP gain

Baseline

29.92

0.3010

7 m 19 s

 

L2-GCN

35.41

0.3744

8 m 01 s

+0.5 %

Back to article page