I'm going to claim this as good enough for two reasons.
1) We're expecting a child any time now, so I expect to be heavily
distracted from this project for a few months.
2) I had someone volunteer to help. After trying these on a couple
books he decided that the results were good enough to use and that
further training wasn't necessary for his books.
I will undoubtedly add more training pages at some point, but this set
is good enough for a fairly large set of books.
Inka Weide of PGDP prepared another training image and box for me from
a book with some difficult fonts, so I'm uploading a new fraktur.tgz.
The example image for Issue 62 is the new training image.
I'm seeing error messages from unicharset_extractor I've not noticed
before. I've consequently fixed about a half dozen errors in my
previous box files. These fixes are in the new fraktur.tgz.
On Aug 31, 9:08 pm, "Ray Smith" <theraysm...@gmail.com> wrote:
> I think I know what the bug is. I will investigate next week.
> Ray.
>
> On 8/31/07, La Monte H. P. Yarroll <piggy.yarr...@gmail.com> wrote:
>
>
>
> > Issue 62 remains a problem with this training dataset.
>