OCR-D / ocrd-website

Creative Commons Attribution 4.0 International
24 stars 7 forks source link

workflows: recommend parameters and recipes #104

Open bertsky opened 4 years ago

bertsky commented 4 years ago

Sometimes a word on parameter choices would be helpful. For example,

Also, beyond full-blown workflow recommendations, simple recipes could be discussed, like:

EEngl52 commented 4 years ago

this is an important issue for sure, thanks for pointing it out @bertsky ! I suggest we will continuously amend our user guide with such pieces of information just as we gain further experience with parameter and processor choices. The pilot phase might come in quite handy in this respect

bertsky commented 4 years ago

@EEngl52 you mean the workflow guide, don't you?

If you make a draft PR with how and where to put such info (i.e. parameters and recipes), then I (or others) can comment/review. Or you wait for others to write about their experiences in the wiki. Maybe we should also discuss this in the VC next week. After all, we ideally want to spawn a user-oriented discussion!

EEngl52 commented 4 years ago

I added some template pages to the wiki where such detailed recommendations can be added.

EEngl52 commented 4 years ago

as we decided to add more in-depth recommendations to the Website Wiki: can we close this issue @bertsky ? Or do you want to keep it as a reminder for your first ideas?

bertsky commented 4 years ago

Yes, good idea. I'll try to add the info somewhere in the wiki, and then close here. Hopefully it gets integrated into the workflow recommendations at some point (probably after having a good working evaluation).

kba commented 1 year ago

@bertsky Is this adressed by the workflow-guide-from-wiki mechanism and the Notes sections there?

bertsky commented 1 year ago

Not quite. The original idea was to provide some middle ground between single processor description and full-blown workflows: simple reusable recipes for special tasks (e.g. how to segment handwriting, how to detect and segment tables, how to do multi-OCR alignment, how to do OCR model selection, how to do cropping with or without facing pages, when and how to do deskewing and dewarping, how to combine segmentation from various tools, how to extract training data suitable for segmentation or for OCR). I'm afraid we don't have that yet, despite some supplementary pages in the Wiki.

Perhaps we can keep this open for our current effort to collect workflow experiences, and try to work this into the WF Guide and (if necessary) additional pages (which we could then link to on the website)?

kba commented 1 year ago

Perhaps we should revisit the original idea of the "OCR-D cookbook" with "recipes" for common tasks and problems like the ones you mentioned. The user guide is too high-level for those, the workflow guide too low-level.

Perhaps it would make sense to combine this with the FAQ (cf. #32). We could have a docu sprint where we collect and answer common questions and based on that decide which question complexes merit a more in-depth analysis? And perhaps have those documents (FAQ and cookbook) live in the Wiki with integration into the website, like the workflows.md mechanism.