Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Repo info
Activity
    Xavier Riley
    @xavriley
    Hello?
    I'm looking for a way of reliably extracting single lines from tables using something like clstm or ocropus. I think I want to just restrict the page segmentation to the Y axis and assume that X is always full width. Is there a way of achieving this?
    mous16
    @mous16
    Good morning to everyone
    I'm looking for an OCR engine capable of eadind a tax form. Is this the right project? Are there some c++ API documentation?
    Pan He
    @BestSonny
    Is there anyone who knows to read and write for the Net with protobuf?
    Sébastien Testeau
    @moskiteau
    hello
    I am trying to install clstm on ubuntu 14.04 and I have errors. I saw a closed bug about it, but it doesn't help
    clstm_proto.cc:18:22: fatal error: clstm.pb.h: No such file or directory
    don't be shy :smile:
    Sébastien Testeau
    @moskiteau
    well, nevermind, I found out the error
    Sébastien Testeau
    @moskiteau
    I am training with run-uw3-500
    it's been training for 12 hours, when does is stop? Do I need to stop it or it will run indifenetly?
    also, did you guys see «nvidia digits»? Can you use it with clstm?
    Shuai Wang GitHub
    @ShuaiGitHub
    @moskiteau Hi,Sébastien, I am also trying to install clstm on Ubuntu 14.04. Do you mind sharing the installation guide? Thanks for your help!!!!
    Tom
    @tmbdev
    Usually, you train for several days and decide when to stop based on a test set
    Johannes Baiter
    @jbaiter
    @kba I could reproduce your segfault in a jessie vm
    it's weird, it only happens if clstm is installed into the system-wide python environment
    inside of a virtualenv it works like a charm
    Konstantin Baierer
    @kba
    @moskiteau @ShuaiGitHub The prerequisites for 14.04 are now in the README (better late than never): https://github.com/tmbdev/clstm#prerequisites
    Johannes Baiter
    @jbaiter
    @kba I found the reason :-)
    it's pillow
    virtualenv had 3.4.2, jessie has 2.6.1
    Johannes Baiter
    @jbaiter
    pushed a fix, no more segfault
    Konstantin Baierer
    @kba
    I can confirm the fix works.
    Konstantin Baierer
    @kba
    And something is seriously wrong with Python paths in this box, usr/lib overriding usr/local/lib, 2.7 modules importing 3.4 modules, it's mayhem and annoying and unproductive, so I've stopped. I can confirm that it works with pip-installed pillow in Python 2.
    Johannes Baiter
    @jbaiter
    :-)
    Johannes Baiter
    @jbaiter
    @kba do you think the training script belongs into clstm? I'm not so sure, since there's already the C++ CLI and the run_uw300 script for demonstration purposes
    on the other hand, maybe we could just throw it the less than idiosyncratic command-line utils and replace them with nice python-based ones
    i have some code for live-plotting train/dev errors with bokeh
    Konstantin Baierer
    @kba
    Mostly, I just wanted something with a cleaner CLI, now. It's not much more than an extended version of the run_uw3 test script.
    Johannes Baiter
    @jbaiter
    s/throw it/replace
    Konstantin Baierer
    @kba
    ?
    Johannes Baiter
    @jbaiter
    no offense, @tmbdev
    i had mix between a typo and a too harsh expression, it was intended as an edit
    Konstantin Baierer
    @kba
    I wouldn't replace the original tools, more like "augment" them :-) To train directly or as example python code to build your own things with it. Mostly I was thinking for testing and CI with all values configurable.
    I would keep the original CLI as they are, since people might be using them in their setups, no need to break those.