Bretislav Hajek

Yes, right now it supports only English alphabet.

Hi, it depends on what model you want to train? For example, `word_classifier_CTC.ipynb` doesn't need the gaplines. The gapline data are necessary only if you want to extract individual letters...

What version of OpenCV do you use? It looks like that in latest version they changed the API and know it works as: ``` contours, hierarchy = cv.findContours(image, mode, method[,...

Please take a look into `ocr_evaluator.ipynb` where is CTC model being used and into `OCR.ipynb` which sums up the whole process (but with a little bit different model).

Training of `CNN-CG` model (it is bit inconsistent with naming). Is done in [`GapClassifier.ipynb`](https://github.com/Breta01/handwriting-ocr/blob/master/GapClassifier.ipynb). But that isn't the best model for separating letters in word. I would recommend you to...

That's not easy to say. Best way is to take some sample text (for which the algorithm fails) and evaluate individual steps. Ideally, you should evaluate each step as if...

This looks like you might run the training script twice which resulted in double initialization of variables. I am planning on updating the scripts, so that it supports retraining and...

I am not sure, but I think the data are same they're just split into different chunks (words, lines, whole forms).

I didn't update the online stored models, but this model wasn't very good. Try replacing it with `models/word-clas/en/SeqRNN/Classifier.meta`

I see that I forgot to include further details on how to extract the data. I guess that you already downloaded the data. Placed them according to the instructions. Then...