@jeremyphoward ‘s and http://fast.ai ‘s influence I’m sure ;)
-
That's only partially true: for discriminative models yes, but there's no inherent reason why convolutions should equals downsampling (for generative models for instance).
*Of course* you can design architectures that don't do downsampling (just don't include maxpooling or strides, and use "same" padding everywhere). And there are use cases for it. I'm just talking about every pre-trained architecture available out there.
-
-
This is completely unviable if you try to do image classification on relatively large inputs, btw.
-
Sure, for classification problems, but not necessarily for language or image modelling or generation where you'd actually prefer to keep image sizes relatively stable in the network, or even enlarge them (eg super resolution).
- Show replies
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.