Multiscale Meets Spatial Awareness: An Efficient Attention Guidance Network for Human Parsing

Table 2

Human parsing performances with our network and ten state-of-the-art methods on ATR [14] test set.

Method	Avg. F1 score
Yamaguchi et al. [40]	41.80
Paper Doll [2]	44.76
M-CNN [41]	62.81
ATR [14]	64.38
DeepLab-v2 (VGG16) [10]	73.53
Attention (VGG16) [17]	77.23
DeepLab-v3 + [42]	76.97
Co-CNN-v1 [18]	80.14
TGPNet [15]	81.76
Co-CNN-v2 [43]	85.36
AG-Net (VGG16)	83.59
AG-Net (DenseNet-121)	85.46