Split layer (split the output of previous layer channelwise into multiple output variables)

[Khurram5]

I need to perform following operation (PyTorch example)

prediction = x.view( batch_size, num_anchors, 85, grid_height, grid_width).permute(0, 1, 3, 4, 2).contiguous()

Basically it is transforming a [1x255x13x13] tensor into [1x3x13x13x85] tensor. Is there any way I can do this operation using TensorRT Python API?

[NVES]

you should be able to do this with a shuffle (possibly you might need more than 1)first you have to reshape to (1,3,85,13,13) and then transpose. Looks like just one shuffle will do it.