do you get good result ever?

[mengzhibin]

I have seen another code about changing cnn feature to lstm as below m = Permute((2, 1, 3), name='permute')(m) m = TimeDistributed(Flatten(), name='timedistrib')(m) but yours is # CNN to RNN inner = Reshape(target_shape=((32, 2048)), name='reshape')(inner) # (None, 32, 2048) inner = Dense(64, activation='relu', kernel_initializer='he_normal', name='dense1')(inner) # (None, 32, 64) I think the reshape function is not correct, So could you tell if you have ever get good result? thank you very much.

[tuanphan09]

I builded my model for captcha recognition base on his model and the result was pretty good. You can vertify it in my project: https://github.com/tuanphan09/captcha-recognition

[Bao150297]

@tuanphan09 Your project is amazing. I wanna train a model used for recognize license plate and have a question about dataset. I read that the input of this model has size 128*64 but my collected data has a variety of size and ratio. Can I use them for training this model? Does it effect to the accurate?

[tuanphan09]

Yes, you can. You have to resize your dataset. It would effect the accurate a little but you can deal with it by data augmentation: shift, zoom, rotation,...Here is how you do it: https://github.com/tuanphan09/captcha-recognition/blob/master/training.py#L44