wushilian / stn_cnn_lstm_ctc_tensorflow Goto Github PK

Is there any STN loss to constraint the transformation angle？I didn't find it in your code. If there is no one, how do the net know the how to transfer a picture?

I'm a newbee here.Ask some stupid questions haha:
1.what dataset did you use?
2.how's the accuracy on svt dataset?
3.How does Densenet perform in this project,better than traditional CNN? Is Densenet hard
to train?

If STN is used, it does not converge.
Is there a good tip to converge?
and
Have you tried evaluating IIIT5k dataset using the original CRNN without using STN?
I implemented it with tf, but it did not perform like the paper.

Hello，I am training this model and I found it hard to converge. Could you tell me how many pictures to train this model and the number of iteration? And what is syn90k, where can I download it? If possible, could you update your final checkpoint. Thank you very much

您好，我在CNN前加入STN后，原本能训练的网络却一直不能训练，就一直卡在读数据的地方，您有遇到过这种问题么？

不太明白怎么使用，可以详细说明一下吗，感谢，我有点小白，看不太懂