search

FIBLAB / DeepSTN

Codes for AAAI 2019 DeepSTN+: Context-aware Spatial-Temporal Neural Network for Crowd Flow Prediction in Metropolis
63 stars 23 forks source link

ConvPlus求助 #2

Closed czisok closed 5 years ago

czisok commented 5 years ago

您好，拜读了您的论文和实验代码，对论文提到的ResPlus中的ConvPlus操作不是很理解，首先论文中提到使用全连接建模long-range spatial dependence，代码中使用的是卷积核为pooling后map的宽高，通道数为plusHW的Conv2D卷积，请问是为什么呢？

vonfeng commented 5 years ago

As https://github.com/FIBLAB/DeepSTN/blob/master/BikeNYC/DeepSTN_network/DeepSTN_net.py#L50 shows, we apply this conv operation to model the relations between any two regions(pixels) in the city, which can be regarded as a fully connected network.

czisok commented 5 years ago

Thanks! Got it

As https://github.com/FIBLAB/DeepSTN/blob/master/BikeNYC/DeepSTN_network/DeepSTN_net.py#L50 shows, we apply this conv operation to model the relations between any two regions(pixels) in the city, which can be regarded as a fully connected network.