Throughout this tutorial, we have guided you step-by-step on how to effectively use Arkindex and Callico for automatic document processing.
You learned how to import your documents into Arkindex and annotate them in Callico for two key document processing tasks: document layout analysis (DLA), and handwritten text recognition (HTR). You also explored how to bring these annotations back into Arkindex to train models for segmentation using YOLO, as well as for text recognition using PyLaia. Finally, we covered how to run these trained models on new data within Arkindex and export the results in Page XML format.
In this section, we will discuss the limitations of the trained models and how they can be improved.
The models trained in this tutorial are based on annotations from a relatively small dataset, as we annotated only 100 pages. While this is sufficient for demonstration purposes, the limited data may impact the performance and accuracy of the models.
Specifically, models trained on small datasets may struggle to recognize diverse layouts or handwriting styles, and their ability to generalize to new documents might be limited.
The models trained in this tutorial could be improved in several ways:
As experts in the field, we can help you design, train, and optimize models designed to your specific needs.
Thank you for following along with this tutorial. We hope it has provided you with the knowledge and tools necessary to fully utilize Arkindex and Callico for your document processing needs. We encourage you to continue experimenting with these platforms. And please, contact us if you have any questions!
The datasets are published, in the right format, on HuggingFace:
The models are also published on HuggingFace:
Head over there to learn how to use them outside the scope of this tutorial.