The parameter in the CNN,RNN is just copy from your paper.And since the frame of torch and pytorch is same, i believe that the code write in pytorch must have the same performance as your code.
But when I testing in the iiit5k,the accuracy is just 65%.I set the learning-rate into 0.001,and using adadelta.
Can you tell me how i can change that or which problem i might have maked
Thank you sincerely
@bgshih
There is another code using pytorch with the same process as your paper
https://github.com/meijieru/crnn.pytorch
The parameter in the CNN,RNN is just copy from your paper.And since the frame of torch and pytorch is same, i believe that the code write in pytorch must have the same performance as your code. But when I testing in the iiit5k,the accuracy is just 65%.I set the learning-rate into 0.001,and using adadelta. Can you tell me how i can change that or which problem i might have maked Thank you sincerely @bgshih