Does CoreML support CTC (connectionist temporal classification) layers?

Hello,


I'm attempting to convert a TensorFlow model to CoreML. The final step in the TensorFlow model uses a CTC beam decoder to determine sequences of characters given a sequence of logit predictions from a set of RNN nodes. However I can't see any reference to an implementation of a CTC layer in CoreML.


Is it supported somehow, or will I have to implement a CTC beam search decoder in Swift as a custom layer?


Cheers

-Thom.

Up vote post of Thom M Down vote post of Thom M
1.3k views
Posted by
Thom M
Reply
Add a Comment

Accepted Reply

There is no such thing in Core ML. If it's the final step, you might as well write it in Swift using the output of the Core ML model instead of using a custom layer.

Posted by
kerfuffle
Up vote reply of kerfuffle Down vote reply of kerfuffle
Post marked as solved
Add a Comment

Replies

There is no such thing in Core ML. If it's the final step, you might as well write it in Swift using the output of the Core ML model instead of using a custom layer.

Posted by
kerfuffle
Up vote reply of kerfuffle Down vote reply of kerfuffle
Post marked as solved
Add a Comment