Rewrite batch question

[Zergratino]

Hi, I'm currently rewriting code for batch inference. Now I encounter some problems. I already rewrote the RPN layers to adapt batch images. However, when I run net.forward(), the result is not consistent. For example, if batch size = 2, net.blobs['rois'].data.shape = (2, 300, 5), and shape of boxes derived from rios is (2, 300, 4) which is consistent with 1 image situation. But the result of output is not like this, blobs_out = net.forward(**forward_kwargs), blobs_out['bbox_pred'].data.shape = (2, 8), it seems to miss one dimension. Any hints or suggests for this condition? Thanks a lot!