Separable convs have been around since 2014, but they're still almost completely absent from convnet design.
New paper: "Deep learning with separable convolutions". https://arxiv.org/abs/1610.02357 - exploring what's next in convnet design after Inception.
-
-
-
But in the future, we will build convnets entirely out of separable convs. They will come to replace regular convolutions in most cases.
- Show replies
New conversation -
-
-
Very nice! This seems likely to be especially useful in the smaller data regime.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Interesting that removing non-linear activations between operations achieved best performance. This has also been my experience.pic.twitter.com/23zF7aIBXQ
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
curious to follow this conversation, maybe you can share later?
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
I have to admit, you win in the naming category. "Xception"
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Do you have the Keras implementation of the architecture somewhere?
-
we will release code and weights files within a few weeks.
End of conversation
New conversation -
-
-
is there an available code of this work?
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Top
#AI New paper: "Deep learning with separable convolutions". https://topdata.news/new-paper-deep-learning-with-separable-convolutions-exploring-whats-next-in-convnet-design-after-inception/ … via@fchollet#architecturepic.twitter.com/tEykPXbgFk
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
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.