Age | Commit message (Collapse) | Author |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
available to Pipeliner
|
|
|
|
lspipeline as there are some hard to debug issues in concurrency version
|
|
The -prefix option is useful to us.
Previously only a .jpg for page number 100 was retreived, which
failed if the book had fewer (or unusually named) pages, and also
didn't provide a corresponding .hocr at all (bug introduced with
48958d2). Using 'best', which is (effectively) randomly sorted,
provides a guaranteed to exist page, and a random one at that.
|
|
ListObjectWithMeta for single file listing, so we can still be as fast, but do not have a misleading api
|
|
|
|
single results from ListObjects requests
|
|
books starting with "1"
|
|
|
|
|
|
already has a hocr directory in it will work
|
|
after the book being processed
|
|
txt-ified on windows
|
|
|
|
|
|
This is important as if a book is added which has already been done,
then an analyse job will be added every time a page is OCRed, which
will clog up the pipeline with unnecessary work. Also if a book was
added with the same name but differently named files, or a different
number of pages, the results would almost certainly not be as
intended.
In the case of a book really wanting to be added with a particular
name, either the original directory can be removed on S3, or "v2"
or similar can be appended to the book name before calling
booktopipeline.
|
|
queue states
|
|
|
|
|
|
they're more user-friendly
|
|
|
|
|
|
some error output)
|
|
TESSDATA_PREFIX accordingly
|
|
|
|
|
|
|
|
|
|
only use 0.1,0.2,0.3
|
|
minimal.
|
|
called rescribe
|
|
No functionality changes, but this should make it easier to make custom
builds using the pipeline in slightly different ways.
|
|
just be stopped after a period, rather than the whole computer shut down
|
|
(failing) log saving, mail sending, and removing erroneous references to AWS
|
|
This ensures that bookpipeline will still work even if TESSDATA_PREFIX has
been set to a directory without configs in it.
|
|
arguments, even if all are strings
|
|
|
|
|
|
There was no reason not to do this with wipeonly as well, and sure enough a
single broken PNG image in a wipeonly task would cause the queue to exponentially
fill as happened previously.
|
|
|
|
|