Framework for mixing Comics datasets, existing and new, with dense and curated annotations.
In the main repository, there will be released:
- instruction to get the datasets (DCM772, comics100, PopManga, eBDtheque, Manga109)
- main code for unifying existing annotations
- converters to UCA
- converters to/from CVAT, COCO, YOLO
As part of the following code, we will release:
- detection annotations (for train split) for the above datasets (curated by authors)
- code and weights for Faster R-CNN, YOLO finetuned models
- code to evaluate the above models, and the existing Magi, DASS, and GroundingDINO
- evaluation server for performance evaluation :)