Robert Sachunsky

Sorry, don't have time to repair this. And as you can see [here](https://github.com/OCR-D/ocrd_anybaseocr/issues/82) and imagine [here](https://github.com/OCR-D/ocrd_anybaseocr/issues/81), block segmentation is not worth investing IMO. We have already been thinking of peeling...

How about changing https://github.com/OCR-D/core/blob/f0e02361f31e7542bcea9a05e316ea161408cdd0/ocrd_utils/ocrd_utils/str.py#L177 into ```python err = ValueError("Not a valid JSON file or string: '%s'" % value) ```

> * link to Quiver from the Workflow Guide It seems Quiver is linked **nowhere** on the website (not even the sitemap). Should I open a separate issue for this...

A workaround is to insert a limit into the current number of `alignments` as a second (non-terminal) case in `backtraceFrom()`: for example, ```python elif len(alignments) > 100*max(f.shape): return # enough...

Good idea! Would you be willing to submit a PR for this (and similar stumbling points)? > Do the `$XDG_` variables matter here at all? One could of course set...

> I think the Deployer agent should be separated by the Processing Server in the [OCR-D System Architecture](https://github.com/OCR-D/spec/blob/master/web_api.md#suggested-ocr-d-system-architecture). On an implementation level, it is already separated. Makes the code easier...

Yes, I have seen this as well. I am also puzzled why despite the `fatal` qualifier, it does go on to update recursively.

I concur with @kba. Adding: * conflicts with outstanding PRs, making it harder to finalize and catch up IMO we don't need these cosmetics, least now.

@z160896 to convert a checkpoint to a traineddata file, just do: lstmtraining --stop_training --continue_from path/to/checkpoint --traineddata path/to/proto.traineddata --model_output path/to/trainedata In your case, that should be: lstmtraining --stop_training --continue_from data/hwmodel/tessdata_best/hwmodel_19.699000_9449_9800.checkpoint --traineddata...

> There's an [LSD implementation in OpenCV](https://docs.opencv.org/3.4/db/d73/classcv_1_1LineSegmentDetector.html) which may represent a possible alternative. Yes – the problem is that it's not nearly as good, and has next to no parameters...