In a Halcon deep OCR use-case for handwritten text, I am having problems with the OCR detection phase.
Considering the low quality of my images the recognition of the text is quite acceptable after training. (using deep_ocr_recognition_training_workflow.hdev)
But no matter I use any of the pre-trained model or my own trained model, the detection remains the same. And it is not a matter of setting detection_min_word_score or detection_min_character_score because based on score maps some text is completely ignored and can't be catchd by lowering min scores as suggested here.
In the documentation for set_deep_ocr_param (get_deep_ocr_param ) it is stated that a parameter "detection_model" can be passed for detection mode. But I can't find how any clue what sort of data learning tool project can be used for this or how the model should be trained.
I mean the parameter detection_model is there. So where is the information on training such model?
I have tried passing the trained model with deep_ocr_recognition_training_workflow.hdev as the detection model but it detects the model type as recognition and shows an error that a model of detection type should be used.