omni-us / research-seq2seq-htr Goto Github PK

Thank you very much for sharing your code.
I am trying to reproduce your results on the IAM database. The training is working fine and the loss is decreasing. However, the attention weights stay at 0 during training.
Do you have an idea of what may be causing this issue ?
Here is an example of a test image after ~20 epochs.

Hi,
When running run_test.sh file I am getting this error. I guess it is coming from the 1149 lines of the htrsh.inc.sh file. The only change I made in htrsh.inc.sh is that I replace gwak to mwak at 1137 line because gwak was not found.
warning: no ground truth for m03-013-01,189
warning: no ground truth for m03-013-02,189
warning: no ground truth for m03-013-03,189
warning: no ground truth for m03-013-05,189
warning: no ground truth for m02-048-00,191
warning: no ground truth for m02-048-01,191
warning: no ground truth for m02-048-03,191
warning: no ground truth for m02-048-04,191
warning: no ground truth for m02-048-05,191
.
.
.
.
for all the files.
Can you help me in this?

Hi,
First of all thank you for sharing this code with us.
I am facing a Permission denied error while executing shell file ./tasas_cer.sh in a main_torch_latest.py in a line 312.
res_cer = sub.Popen(['./tasas_cer.sh', gt, decoded], stdout=sub.PIPE)

I also add shell=True but got the same error again.
I also tried to change the permission of that file by executing python command
os.chmod('tasas_cer.sh', stat.S_IXOTH)
but unfortunately I got the same error again.
I am running this code on google colab.
Can you help me out please?

hello, in order to Calculating the CER , i run ''python3 pytasas_words.py 1000 si'',the result as follow :

can you help me to solve the problem

thank you

Hi in

Hi , first of thanks for creating this repository and sharing it with community ..I had some confusion on reading the Research paper & code.
In page 5 of your research paper..under content based attention , you have mentioned:

-hi as the hidden state of the encoder at the current time step i∈ {0,1,...,N-1}
but shouldn't hi be the output of the encoder

-st as the hidden state of decoder at current time step t∈ {0,1,...,T-1}, where T is the maximum length of decoding characters
I believe this is implemented at line 48 of seq2seq.py How come the t∈ {0,1,...,T-1} , how can t be 0 ,as there will never case where t is 0 .