The benchmarks/detection
folder contains the code for the following models:
Faster R-CNN
DASS
YOLOv8
Grounding Dino
Magi
SSD
(to be added)
The benchmarks are run on the validation split of the CoMix
dataset, which comprehends:
- Comics-style:
- comics100 (val, test, dev[todo])
- DCM (val, test)
- eBDtheque (val, test)
- Manga-style:
- PopManga (val, test)
- Manga109[todo] (val, test, dev)
For these datasets, follow the instructions in the CoMix/README.md and then datasets/README.md files to preprocess and convert the data.
Here you can find the instructions for running the models generation and evaluating the predictions evaluation.