is the lstm2d the true mdlstm ?

[Duum]

I see you have changed the mdlstm to lstm2d, but I think the lstm2d implemented by tensorflow does not consider the both dimension as introduced by Alex Graves. Have you tested its accuracy?

[johnsmithm]

both has the accurancy about 15% label rate error, but lstm2d is faster because in the mdlstm tf.while_loop has parallel_iterations=1. lstm2d still has the information from both dimensions, but the information from another dimenstion is input, but not passed through a gate as in the original multidimentional lstm

[Duum]

Do you have the interest to implement this

Joint Line Segmentation and Transcription for End-to-End Handwritten Paragraph Recognition

?

[johnsmithm]

It is interesting

[Duum]

I have complete a version, but it not works well. If you want to implement it, I'm willing to help.

[johnsmithm]

close for non activity