Created by: pengwangucla

It is not convenient to reshape the output from paddle.infer for any inter output

For example:

conv = paddle.layer.img_conv(...)
out = paddle.infer(output_layer=conv,
                    parameters=parameters,
                    input=test_data,
                    feeding={'image1': 0})

then I need to manually give the height and width for the 'out' in order to view it as a 4 D image tensor.

Is it possible to add height and width field in the LayerOutput object in paddle ? So the height and width of an output is automatically reshaped.

out = out.reshape((batchsize, conv.num_filters, conv.height, conv.width)) @jacquesqiao