Skip to main content

Advances, Systems and Applications

Table 4 Comparisons with state-of-the-arts and ablation study

From: Visibility estimation via deep label distribution learning in cloud environment

Methods Label type FROSI FRIDA RMID
VGG16 [55] absolute 33.0% 26.4% 39.8%
ResNet [56] absolute 26.3% 25.7% 30.2%
CNN+GRNN [7] absolute 23.8% 29.1% 35.6%
DHCNN [57] absolute 28.6% 29.2% 41.9%
VisNet [8] absolute 21.4% 22.9% 27.5%
Relative CNN-RNN [16] ranking 14.5% 13.8% 18.3%
Ours (without CNN) distribution 23.3% 30.0% 33.6%
Ours (without RNN) distribution 14.6% 16.7% 18.1%
Ours (without LDL) absolute 14.0% 15.1% 16.3%
Ours (without Fusion) distribution 12.1% 13.7% 15.8%
Ours (average fusion) distribution 11.3% 11.4% 13.9%
Ours (voting fusion) distribution 11.5% 12.0% 13.3%
Ours (max fusion) distribution 10.8% 9.9% 14.7%
Ours distribution 9.7% 8.9% 11.6%
  1. The best performance indicator is marked as bold