Text-line detection & extracion [segmenting pages into lines]

[ghost]

@ChWick Thank you for your hard work.

• Any updates regarding your progress in implementing your Page Segmentation FCN network to Calamari?

• Will you include CPU mode for training & prediction?

[ChWick]

@mrocr

• Unfortunately there are no updates on the FCN, yet
• CPU will be used by default for training and prediction if no valid GPU is available

[ghost]

We are still hoping that you release a text line detection and extraction implementation in calamari in the future.
Thank you.

[ghost]

@ChWick Is there any tool available at GitHub that you recommend using for training text-line detection and extraction?

[ChWick]

@mrocr Maybe the line segmenter of T. Breuel https://github.com/NVlabs/ocroseg is what you are looking for.

[ghost]

The developer of Kraken has noted to me that ocroseg doesn't converge, even when using the same uw3 dataset that T. Breuel used.
Source of conversation: mittagessen/seg#4

[madisonmay]

Hey there @ChWick. The README states that you intend to eventually move the ocropy line segmentation step into calamari. Is there a branch where this work is in progress that I could potentially build off of?

[kba]

@madisonmay You can also use kraken's segmentation which handily produces JSON of bounding boxes (https://github.com/mittagessen/kraken/blob/master/docs/advanced.rst#page-segmentation-and-script-detection). Ocropy has two PR (ocropus-archive/DUP-ocropy#281 and ocropus-archive/DUP-ocropy#283) which also provided such coordinates JSON.