The Comics Datasets Framework

Framework for mixing Comics datasets, existing and new, with dense and curated annotations.

Structure

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 :)