Detectron2 v2 model
This repository hosts version 2 of our trained Detectron2 model (sucessor to previous trained model), that can detect segments from digitized books.
The following classes are supported:
- Illustration
- Stamp
- Initial
- Other
The model is based on faster_rcnn_R_50_FPN_3x and is fine-tuned on 8.027 manually annotated images, resulting in 5.363 annotated segments.