Feat/flex line dirs

[mikegerber]

This adds more flexibility w.r.t. evaluating directories of line texts.

---

• Test dinglehopper

  • Check support for "plain-encoding"

• Test dinglehopper-line-dirs

  • Check support for "plain-encoding"

• Test dinglehopper-extract

  • Check support for "plain-encoding"

• Test dinglehopper-summarize

• Test ocrd-dinglehopper

  • Check support for "plain-encoding"

• Update docs w.r.t this feature

  • dinglehopper-line-dirs --help
  • README.md

• Review Unexpected UTF-8 problems #123

[mikegerber]

Needs more testing, converting to draft for now.

[mikegerber]

I've added a check list above to go through the various CLIs and test them. Because this also adds support to specify a plain text encoding. I've also added this to the check list.

[mikegerber]

dinglehopper CLI works fine.

[mikegerber]

Ha, dinglehopper-extract doesn't have --plain-encoding yet (but complains about auto-detecting) → Needs a fix.

[mikegerber]

Ha, dinglehopper-extract doesn't have --plain-encoding yet (but complains about auto-detecting) → Needs a fix.

Fixed in 14a4bc5.

[mikegerber]

dinglehopper-summarize does not read any input files, so it's not relevant here. Should be tested in #112.

[mikegerber]

dinglehopper-line-dirs behaves as expected!

[mikegerber]

Manual test of ocrd-dinglehopper also correctly warns about autodetecting the plain text encoding + has the option to give an explicit encoding.

❯ ocrd-dinglehopper -P plain_encoding utf-8 -I GT-TXT,OCR-D-OCR-TESS -O DINGLEHOPPER-FROM-TXT2

Don't see how to stick the information about the plain text encoding into the METS file - that could be an improvement over this. Maybe @bertsky has an idea?

(I see comparing to txt GT as useful in some cases, e.g. when working with corpora where only the text is available but no PAGE/ALTO.)

[mikegerber]

The help text of dinglehopper-line-dirs looks ready.

❯ dinglehopper-line-dirs --help
Usage: dinglehopper-line-dirs [OPTIONS] GT OCR [REPORT_PREFIX]

  Compare the GT line text directory against the OCR line text directory.

  This assumes that the GT line text directory contains textfiles with a
  common suffix like ".gt.txt", and the OCR line text directory contains
  textfiles with a common suffix like ".some-ocr.txt". The text files also
  need to be paired, i.e. the GT filename "line001.gt.txt" needs to match a
  filename "line001.some-ocr.txt" in the OCR lines directory.

  GT and OCR directories may contain line text files in matching
  subdirectories, e.g. "GT/goethe_faust/line1.gt.txt" and
  "OCR/goethe_faust/line1.pred.txt".

  GT and OCR directories can also be the same directory, but in this case you
  need to give --gt-suffix and --ocr-suffix explicitly.

  The GT and OCR directories are usually ground truth line texts and the
  results of an OCR software, but you may use dinglehopper to compare two OCR
  results. In that case, use --no-metrics to disable the then meaningless
  metrics and also change the color scheme from green/red to blue.

  The comparison report will be written to $REPORT_PREFIX.{html,json}, where
  $REPORT_PREFIX defaults to "report". The reports include the character error
  rate (CER) and the word error rate (WER).

  It is recommended to specify the encoding of the text files, for example
  with --plain-encoding utf-8. If this option is not given, we try to auto-
  detect it.

Options:
  --metrics / --no-metrics  Enable/disable metrics and green/red
  --gt-suffix TEXT          Suffix of GT line text files
  --ocr-suffix TEXT         Suffix of OCR line text files
  --plain-encoding TEXT     Encoding (e.g. "utf-8") of plain text files
  --help                    Show this message and exit.
  ```

[mikegerber]

README.md now points to dinglehopper-line-dirs --help for the special case of mixed GT/OCR line text directories.

[mikegerber]

I've added a test for plain text files with BOM.