search

garyzhao / SemGCN

The Pytorch implementation for "Semantic Graph Convolutional Networks for 3D Human Pose Regression" (CVPR 2019).
https://arxiv.org/abs/1904.03345
Apache License 2.0
467 stars 78 forks source link

About the chanel-wise #36

Open ZJJTSL opened 4 years ago

ZJJTSL commented 4 years ago

i can't understand the chanel-wise implementation, file "sem_ch_graph_conv.py" and "sem_graph_conv.py", what's the difference between them. And i wonder the size of "W" in file "sem_ch_graph_conv.py" line 19, how to understand its first dimention of 2. Looking forward your guidance ,thanks!

garyzhao commented 3 years ago

Hi @ZJJTSL ,

Thanks for your interest in our work!

"sem_graph_conv.py" is the implementation according to Eq. (2) in the main paper, while "sem_ch_graph_conv.py" is the channel-wise implementation according to Eq. (3). We use two weights for the node itself and its neighbors, respectively, so that's why the first dimension is two. For more details, please see https://arxiv.org/pdf/1904.03345.pdf Eq. (9) on page 12 (Section A.2.). Thanks!

Best, Long