Word/glyph segmentation coordinates are broken

[mikegerber]

There is a cluster of bounding boxes elsewhere on the page, so I think the coordinates are broken probably.

[mikegerber]

👀 @VChristlein

The glyphs are there, coordinates seem fine too (casual inspection), but not displayed in Page Viewer. I'll investigate more thoroughly next week.

[mikegerber]

ocrd-segment-extract-pages -I OCR-D-OCR-CALAMARI -O EXTRACT-PAGES -P plot_segmasks '["page", "region", "line", "word", "glyph"]' -P plot_overlay true

This produces the same problem as Page Viewer, so I guess the problem is with ocrd_calamari ;-) (Image scaled for GitHub upload)

[mikegerber]

The code has been broken since we try to process all lines of a region at once. We are always using the y coordinates of the last line in the text region 😅

[mikegerber]

This is fixed in the latest release.

[mikegerber]

Reopening: I'd like a simple test for this as I don't routinely check word coordinates. I imagine a simple document and a rough check if the word coordinates are plausible.

[kba]

You could try https://github.com/OCR-D/assets/tree/master/data/kant_aufklaerung_1784-page-region-line-word_glyph/data as test data.