Exchanging preprocessor

[twerkmeister]

Hi @ChWick @andbue!

Thanks for this amazing project. I am using calamari as part of a data extraction task for tables in mid 20th century documents. Specifically I run calamari on the (single line) cells of the tables and had a really satisfying experience so far and made good progress training and tuning on my growing dataset (12500 cell images by now). I've been digging a bit deeper into the documentation and code and one of the experiments I want to try is turning of the center normalizer preprocessor as it seems to be doing a few things that I am not sure are necessary or helpful for my data. For example, image quality at times is already low, so additional blurring might be hurting. Second, my lines aren't skewed. I deal with skewing before overlaying the table grid onto the page.

I saw this paramater in the docs:

  --data.pre_proc.processors [DATA.PRE_PROC.PROCESSORS [DATA.PRE_PROC.PROCESSORS ...]]

and also this code in another issue

>>> from calamari_ocr.ocr.dataset.data import Data
>>> params = Data.default_params()
>>> list(enumerate(params.pre_proc.processors))
[
(0, CenterNormalizerProcessorParams(modes={<PipelineMode.TRAINING: 'training'>, <PipelineMode.PREDICTION: 'prediction'>, <PipelineMode.EVALUATION: 'evaluation'>}, extra_params=(4, 1.0, 0.3), line_height=-1)), 
(1, FinalPreparationProcessorParams(modes={<PipelineMode.TRAINING: 'training'>, <PipelineMode.PREDICTION: 'prediction'>, <PipelineMode.EVALUATION: 'evaluation'>}, normalize=True, invert=True, transpose=True, pad=16, pad_value=0)), 
(2, BidiTextProcessorParams(modes={<PipelineMode.TRAINING: 'training'>, <PipelineMode.TARGETS: 'targets'>, <PipelineMode.PREDICTION: 'prediction'>, <PipelineMode.EVALUATION: 'evaluation'>}, bidi_direction=<BidiDirection.AUTO: 'auto'>)), 
(3, StripTextProcessorParams(modes={<PipelineMode.TRAINING: 'training'>, <PipelineMode.TARGETS: 'targets'>, <PipelineMode.PREDICTION: 'prediction'>, <PipelineMode.EVALUATION: 'evaluation'>})), 
(4, TextNormalizerProcessorParams(modes={<PipelineMode.TRAINING: 'training'>, <PipelineMode.TARGETS: 'targets'>, <PipelineMode.PREDICTION: 'prediction'>, <PipelineMode.EVALUATION: 'evaluation'>}, unicode_normalization='NFC')), 
(5, TextRegularizerProcessorParams(modes={<PipelineMode.TRAINING: 'training'>, <PipelineMode.TARGETS: 'targets'>, <PipelineMode.PREDICTION: 'prediction'>, <PipelineMode.EVALUATION: 'evaluation'>}, replacement_groups=[<ReplacementGroup.Spaces: 'spaces'>], replacements=None)), 
(6, AugmentationProcessorParams(modes={<PipelineMode.TRAINING: 'training'>}, augmenter=DefaultDataAugmenterParams(), n_augmentations=0)), 
(7, PrepareSampleProcessorParams(modes={<PipelineMode.TRAINING: 'training'>, <PipelineMode.PREDICTION: 'prediction'>, <PipelineMode.EVALUATION: 'evaluation'>}))
]

What I would like to try is to just exchange the center normalizer preprocessor for the basic scale to height preprocessor. But I am not sure how to achieve this. Do I need to define all the preprocessors and their parameters using --data.pre_proc.processors? Would you mind giving me an idea how to reference the classes and their parameters properly? And does my reasoning to exchange the preprocessor have some merit?

Best, Thomas

[andbue]

Hi Thomas, it is not possible to insert another processor without providing the full set of parameters by providing module path and ProcessorParams-class for each of them. It's quite a mouthful:

--data.pre_proc.processors \
calamari_ocr.ocr.dataset.imageprocessors.scale_to_height_processor:ScaleToHeightProcessorParams \
calamari_ocr.ocr.dataset.imageprocessors.final_preparation:FinalPreparationProcessorParams \
calamari_ocr.ocr.dataset.textprocessors:BidiTextProcessorParams \
calamari_ocr.ocr.dataset.textprocessors:StripTextProcessorParams \
calamari_ocr.ocr.dataset.textprocessors:TextNormalizerProcessorParams \
calamari_ocr.ocr.dataset.textprocessors:TextRegularizerProcessorParams \
calamari_ocr.ocr.dataset.imageprocessors:AugmentationProcessorParams \
calamari_ocr.ocr.dataset.imageprocessors:PrepareSampleProcessorParams

[twerkmeister]

Thank you so much @andbue I think that's the starting point I needed 👍 will try it out tomorrow

[bertsky]

This should be way easier to find. The CLI (tfaip's subcommand self-documentation) is not helpful.

Also, this should be the default if passing --train.channels 3...