Note that both confidence voters depend on the loss function used for training a model, while the sequence voter can be used for all models but might yield slightly worse results. The voting algorithm can be changed by the -voter flag. To enable voting you simply have to pass several models to the -checkpoint argument. Calamari also supports several voting algorithms to improve different predictions of different models. The prediction step using very deep neural networks implemented on Tensorflow as core feature of calamari should be used. In the meantime you should use the scripts provided by OCRopus. Modules to segment pages into lines will be available soon. Currently only OCR on lines is supported. Note that you have to activate the virtual environment if used during the installation in order to make the command line scripts available. If you simply want to use calamari for applying existent models to your text lines and optionally train new models you probably should use the command line interface of calamari, which is very similar to the one of OCRopy. It is designed to both be easy to use from the command line but also be modular to be integrated and customized from other python scripts.

Mon calamari cruiser model