Is there a better way to process tensors than looping?

[ryanaleksander]

As of right now I can't find a way to process tensors efficiently as numpy can. I'm deploying a deeplab model onto an Android app, and the output tensor has the shape of [1x21x400x400]. With numpy i would just do np.argmax(out, axis=1), but on Android, I have to loop through the entire thing which is painfully slow.