search

hlzhu09 / Hierarchical-Human-Parsing

75 stars 14 forks source link

Where do you use GNN and convGRU? #4

Open zzw1123 opened 4 years ago

zzw1123 commented 4 years ago

Thanks for your great work! But I find the adjacent matrix parameter is not used in the following code:

for i in range(self.edge_index_num):
            self.part_list_list[self.edge_index[i, 1]].append(self.edge_index[i, 0])

https://github.com/hlzhu09/Hierarchical-Human-Parsing/blob/0a4b26c20159ca7da6374278304535813b4e0907/utils/gnn_loss.py#L19 So I have a doubt about how to build a GNN......

And the convGRU mentioned in the paper is not used either....

hlzhu09 commented 4 years ago

Our model is described in https://github.com/hlzhu09/Hierarchical-Human-Parsing/blob/master/network/gnn_parse.py The whole graph consists of three subgraphs: full body graph, half body graph, and detailed part graph. The convGRU is used as update operation for each node.

zzw1123 commented 4 years ago

Thanks for releasing new code! So in order to train the network, can I directly replace the 16th line in train_baseline.py from from network.baseline import get_model to from network.gnn_parse import get_model?