On 20 July 2015 at 20:40, Tom Morris <
tfmo...@gmail.com> wrote:
> Jeff's answer is probably the most important explanation, but some other
> reasons include:
> - Tess supports more languages
> - Tess is older
> - Tess has a bigger more well developed community (partly because of all the
> other reasons)
> - Tess is higher performance (from a resource utilization point of view,
> last time I checked)
>
> Ocropus is/was pretty much a one-man project and was, as I understand it,
> designed to support his research.
That seems to be true at the moment, but there were a few people
working on it at different stages.
> It also went through a significant
> rewrite as a result of a change in implementation strategy and that
> discontinuity probably didn't help things.
Several rewrites. One of them was based around line-oriented FSTs, and
interestingly enough, similar work was done with Tesseract
(
https://github.com/tesseract-ocr/tesseract/pull/29). The current
version of OCRopus uses LSTM, which is one of the currently
fashionable types of neural network.
--
<Sefam> Are any of the mentors around?
<jimregan> yes, they're the ones trolling you