About the propagation process(rnn_1---rnn_8,w_x1,w_x2,w_y1,w_y2)

[sHaol0w]

Hi danieltan07,
I am currently implementing spatial propagation using pytorch. I have some questions about your spatial propagation code in Class VGG16modified, model.py.

1. rnn_1 is for left to right propagation and use w_x1 weight. And rnn_2 is for right to left propagation. Why does rnn_2 use w_x1? Maybe it should use w_x2?
2. What do rnn5--rnn8 stand for? Why do we do one more time propagation after rnn1--rnn4?
   Hope you can help me. Thanks!!

(And I have fixed some troubles when using not square input)