issue about y_pred

[18370029656]

May I ask, in this code, what is the shape of y_pred? Is it [1,1, image size]?

[MungoMeng]

Hi, thanks for your interest in our work!
The shape of y_pred in the loss Lstructure is [batch size, image size, 2], where the 2 is the channel number.
Please refer to the function diffs_with_edge. The first channel [..., 0:1] is the predicted appearance transformation map, while the second channel [..., 1:2] is the edge map.