confused about refined feature map

[coordxyz]

hi, I'm confused about the refined feature map. Should it be h2^att rather than h1^att? In your paper, Equ.10 describes the refined feature h2^att is generated from h1^att, but in section 5.2.4, only the h1^att is used as refined feature and to do final prediction.

This is also conflict with Figure 4 in which the h2^att is the refined feature and used for heatmap prediction.

If h1^att is the refined feature, what does the h2^att used for?

Thank you~

[xizero00]

对于第四个到第8 个hourglass模块而言，他的预测结果并不是传统的方式得到的 而是通过对f这个特征首先进行1x1卷积，得到sp，sp是一个1通道的featuremap，然后这个1通道的featuremap送到M内，得到attention map，\Phi_p（它也是1通道的），然后这个attention map乘以h_1^{att}（它不是1通道，比如256个通道的）得到经过对应于某个关节点的attention map得到的特征，然后再使用一个1x1卷积得到1通道的某个关节点的热力图。