Changing the maximum length of string

[stalagmite7]

@bgshih , I have a dataset where the labels exceed the 26-character max length that the model needs, but I am unable to find where in the code I need to make the change. I have changed the line in config.lua where it assigns maxT=26, but the is a tensor size mismatch during validation because the outputBatch after having passed through the network is of length 26 still. Could you tell me where I need to fix it in the code?

[ibrahimsoliman97]

@stalagmite7 I have tried maxT = 52 but I got error in tensor size mismatch, so I changed the 4th maxpooling window to 1x1 instead of 1x2 . then it is start training successfully but after the training finish , the result was as shown in the screenshot "it just detect the letters in the first or last 26 frames and the other frames return ( - )", this make me very confusing about how the network divide the image in number of frame like 26 or 52.. I hope if @bgshih can a little bit explain this to us.. Thank you

[stalagmite7]

I didn't have time to change the model this time around, so I just made sure my inputs are all within 26 characters.