You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I recently installed ocropus on a server on Amazon's EC2, based on their latest "Amazon Linux 2 AMI", which I believe is based on Fedora (it certainly uses yum). On this, running ocropus-rpred with the --probabilities argument causes a traceback error:
Traceback (most recent call last):
File "/usr/bin/ocropus-rpred", line 276, in safe_process1
return process1(arg)
File "/usr/bin/ocropus-rpred", line 203, in process1
result = lstm.translate_back(network.outputs,pos=2)
File "/usr/lib/python2.7/site-packages/ocrolib/lstm.py", line 776, in translate_back
if pos==2: return [(c, outputs[r,c]) for (r,c) in maxima] # include character probabilities
IndexError: only integers, slices (`:`), ellipsis (`...`), numpy.newaxis (`None`) and integer or boolean arrays are valid indices
I am guessing that it's numpy that is the cause of this issue. Version 1.7.1 is installed.
I found the reason for it is that the maxima array is full of floats (like 12.0), which aren't valid as array indices. It can be fixed by changing the offending line to: if pos==2: return [(c, outputs[int(r),int(c)]) for (r,c) in maxima] # include character probabilities
I've done that in my 'fixprobs' branch, commit ebd462b , and it seems to work fine. There could be more places where maxima using floats is an issue, but I haven't come across any.
The text was updated successfully, but these errors were encountered:
I recently installed ocropus on a server on Amazon's EC2, based on their latest "Amazon Linux 2 AMI", which I believe is based on Fedora (it certainly uses yum). On this, running ocropus-rpred with the --probabilities argument causes a traceback error:
I am guessing that it's numpy that is the cause of this issue. Version 1.7.1 is installed.
I found the reason for it is that the maxima array is full of floats (like 12.0), which aren't valid as array indices. It can be fixed by changing the offending line to:
if pos==2: return [(c, outputs[int(r),int(c)]) for (r,c) in maxima] # include character probabilities
I've done that in my 'fixprobs' branch, commit ebd462b , and it seems to work fine. There could be more places where maxima using floats is an issue, but I haven't come across any.
The text was updated successfully, but these errors were encountered: