If I remember correctly, the lasagna cost functions all assume 1d data, with the category axis last. So you'll need to reshape your prediction and your truth mask to be something like batch,pixel,category.
Argmax returns the argument (the location) of the maximum element, not the maximum element itself. So if you do an argmax across your category dimension you'll get back a multi-valued mask where the pixel values indicate the predicted class. If you want a set of binary masks, probably the easiest way to get it is to split up the multi-valued mask using something like np.where. -- --- You received this message because you are subscribed to the Google Groups "theano-users" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/d/optout.
