Specify epochs?

[daniiki]

Hi, where can you specify the number of epochs for training? Thanks!

[Pay20Y]

Sorry, the epochs can not be specified directlly in this code, please calculate the corresponding number of iterations.

[daniiki]

Thanks for the fast answer. Can you tell me how to compute the recognition accuracy / precision?
I computed the detection accuracy like this:

d_correct_prediction = tf.equal(tf.argmax(f_geometry, 1), tf.argmax(input_geo_maps, 1))
detection_accuracy = tf.reduce_mean(tf.cast(d_correct_prediction, tf.float32))

but the calculation of the recognition accuracy seems to be more complicated, since input_transcription is a sparse tensor.

[Pay20Y]

Hi, you can modify the data generator and make it yield another string of label. I recommend that you refer to the ICDAR official evaluation script.