difference with PSPNet #61
Description
I noticed that in your code of PyramidOCP, you partition the feature map first for OCP and then concat them. Also, the concat order is along the x axis. This operation is more complex than PSPNet pooling operation, and what is the motivation of it ?