I tried to adapt some work built on @danvk for this purpose, i.e. to show probabilities for each letter. It is really just a small hack from my side, but you can have a look at the branch https://github.com/zuphilip/ocropy/commits/probabilities

Is it fast running, why not to use fast framework like Theano, they have
lstm too
On Mar 3, 2016 06:59, "Jason Yang" notifications@github.com wrote:

We are running it on some documents and we need absolute accuracy so we
are using human proofreading, but is there any way to access confidence
level so that we can just examine the ones with low confidence level?

—
Reply to this email directly or view it on GitHub
#87.

thank you, @zuphilip, I am checking that! It is great! By the way, do you know whether the result comes out every time is same or not? I mean if I use same model on same image, is it possible to get different results?

@Sandy4321 It's great to know Theano, but it seems it's just a mathematic library, how does that relate to this?

It is deep learning framework
On Mar 3, 2016 18:11, "Jason Yang" notifications@github.com wrote:

@Sandy4321 https://github.com/Sandy4321 It's great to know Theano, but
it seems it's just a mathematic library, how does that relate to this?

—
Reply to this email directly or view it on GitHub
#87 (comment).

@zuphilip I just tried your version, it is so so so nice! But I have some problem to interpret the probability numbers, for example the ones listed below. what is the first number and second number mean in each line? BTW, It is great to see your work, it's really amazing and useful.

4   21.0    0.998147512334
1   18.0    0.998673532294
6   23.0    0.933822265371
    1.0 0.985126139538
W   56.0    0.995752034066
    0.0 0.669438453921
l   77.0    0.771479722076
l   77.0    0.826818186054
l   77.0    0.862084065346
a   66.0    0.992835549207
m   78.0    0.995445149948
s   84.0    0.982924709589
    1.0 0.986920309815
J   43.0    0.917265339312
    1.0 0.991153504035
W   56.0    0.665979106147
    0.0 0.343662815981
    1.0 0.993414011944
(   9.0 0.988318526414
e   70.0    0.664349017539
)   10.0    0.994918996523
    1.0 0.993739754716
g   72.0    0.981143592017
r   83.0    0.997430796532
o   80.0    0.989413900792

I am no expert on this subject and I just tried out some code. The first number looks to be some index for the representing letter, for example all W are encoded with 56. The first column is calculated from the second column by the code l = network.l2s([p]), but It haven't looked further what this means. Therefore, the really interesting thing is the third column. I have no clue about the calculation. There are some older issues on this, see #25, #47. Maybe that helps.

May you pls share code to try
On Mar 4, 2016 01:45, "Philipp Zumstein" notifications@github.com wrote:

I am no expert on this subject and I just tried out some code. The first
number looks to be some index for the representing letter, for example all
W are encoded with 56. The first column is calculated from the second
column by the code l = network.l2s([p]), but It haven't looked further
what this means. Therefore, the really interesting thing is the third
column. I have no clue about the calculation. There are some older issues
on this, see #25 #25, #47
#47. Maybe that helps.

—
Reply to this email directly or view it on GitHub
#87 (comment).

You really helped a lot, thanks so much @zuphilip

whose code are you asking? @Sandy4321 zuphilip has his code in his repository and he has url above.

super
thanks a lot
which project you mean from
https://github.com/zuphilip?tab=repositories

On Fri, Mar 4, 2016 at 10:43 AM, Jason Yang notifications@github.com
wrote:

whose code are you asking? @Sandy4321 https://github.com/Sandy4321
zuphilip has his code in his repository and he has url above.

—
Reply to this email directly or view it on GitHub
#87 (comment).

Same project: ocropy, different fork, different branch:
https://github.com/zuphilip/ocropy/tree/probabilities

We should put this in the wiki and/or merge it.

yes, but it seems this repository is not maintained any more?

An imroved version is now merged. Example for usage:

ocropus-rpred tests/1175-01002b.png --probabilities