OCR for some images fails with 'Delaunay' object has no attribute 'vertices'

[stweil]

Kraken fails on several images of one of our publications, for example on physical page 10:

(venv3.10) stweil@ocr-02:~/ocr$ kraken --input 1694884104_0010.jpg 1694884104_0010.txt segment --baseline ocr --model austriannewspapers.mlmodel
scikit-learn version 1.2.2 is not supported. Minimum required version: 0.17. Maximum required version: 1.1.2. Disabling scikit-learn conversion API.
Loading ANN /home/stweil/ocr/venv3.10/lib/python3.10/site-packages/kraken/blla.mlmodel	✓
Loading ANN austriannewspapers.mlmodel	✓
Segmenting	✓
Processing ━╺━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━   4% 3/81 0:00:45 0:00:02
[06/30/23 22:13:28] ERROR    Failed processing 1694884104_0010.jpg: 'Delaunay' object has no attribute 'vertices'                                                                                   kraken.py:418

The same error also occurs with other OCR models.

It can also be reproduced with Python 3.11 and newer Torch (requires pull request #518):

(venv3.11) stweil@ocr-02:~/ocr$ kraken --input 1694884104_0010.jpg 1694884104_0010.txt segment --baseline ocr --model austriannewspapers.mlmodel
scikit-learn version 1.2.2 is not supported. Minimum required version: 0.17. Maximum required version: 1.1.2. Disabling scikit-learn conversion API.
Torch version 2.0.1+cu117 has not been tested with coremltools. You may run into unexpected errors. Torch 2.0.0 is the most recent version that has been tested.
Loading ANN /home/stweil/ocr/venv3.11/lib/python3.11/site-packages/kraken/blla.mlmodel	✓
Loading ANN austriannewspapers.mlmodel	✓
Segmenting	✓
Processing ━╺━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━   4% 3/81 0:00:35 0:00:01
[06/30/23 22:30:12] ERROR    Failed processing 1694884104_0010.jpg: 'Delaunay' object has no attribute 'vertices'                                                                                   kraken.py:418

[stweil]

Downgrading from scipy-1.11.1 to scipy-1.10.1 helps:

(venv3.11) stweil@ocr-02:~/ocr$ kraken --input 1694884104_0010.jpg 1694884104_0010.txt segment --baseline ocr --model austriannewspapers.mlmodel
scikit-learn version 1.2.2 is not supported. Minimum required version: 0.17. Maximum required version: 1.1.2. Disabling scikit-learn conversion API.
Torch version 2.0.1+cu117 has not been tested with coremltools. You may run into unexpected errors. Torch 2.0.0 is the most recent version that has been tested.
Loading ANN /home/stweil/ocr/venv3.11/lib/python3.11/site-packages/kraken/blla.mlmodel	✓
Loading ANN austriannewspapers.mlmodel	✓
Segmenting	✓
Processing ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% 81/81 0:00:00 0:00:15
Writing recognition results for 1694884104_0010.jpg	✓

[stweil]

See other report of the same problem in the eScriptorium chat: https://matrix.to/#/!kwucPfyPlAhwGUrUBl:gitter.im/$Zu0JLPGHIjrjX1xWWV82G3Pvws5FS8-tSaTWzfUyKiY?via=gitter.im&via=matrix.org&via=fau.de.

[mittagessen]

Arrgh, sorry. I've tagged a new release (4.3.13) containing a pinned scipy as a hotfix. I'll have a look on the actual reason tomorrow.

[mittagessen]

It's a regression in scikit-image. It is fixed in their latest release but they don't pin their requirements so it's easy to have something like this slip through.