search

emedvedev / attention-ocr

A Tensorflow model for text recognition (CNN + seq2seq with visual attention) available as a Python package and compatible with Google Cloud ML Engine.
MIT License
1.08k stars 256 forks source link

training parameter on Synth 90k #125

Closed roger12337 closed 5 years ago

roger12337 commented 5 years ago

What is the optimal training parameters for Synth 90k? When does it converge?

emedvedev commented 5 years ago

The defaults are fairly optimal, except you might want to increase the batch size (to something like 20) and decrease the number of epochs.

Make sure you adjust max-width, max-height, and max-prediction, too.