search

jonrein / tensorflow_CTC_example

A short example of training a bidirectional LSTM model with connectionist temporal classification
154 stars 52 forks source link

Training images #7

Open joyJZhang opened 7 years ago

joyJZhang commented 7 years ago

hello,i used your model to train myself captcha images,but the loss is always 14,and don't convergence。Can you show me your training images?or tell me how to see them .I want to know whether my pictures are too complicated。Thanks.

jonrein commented 7 years ago

The inputs in the example code are processed features from speech waveforms, not images. If you're trying to use this type of model structure with image data, you'll likely need a few more layers, with possibly a mix of convolution and LSTM layers.

mrgloom commented 7 years ago

Hello, can you specify how to prepair input and output for OCR like tasks?

Related: https://github.com/igormq/ctc_tensorflow_example/issues/20