Active learning for Mask R-CNN in Detectron2

I've created new active learning software that can significantly reduce the annotation effort for training Mask R-CNN in Detectron2. The code can be found here: https://github.com/pieterblok/maskal

There is also a paper publication on arXiv: https://arxiv.org/abs/2112.06586

Enjoy!

Start by reviewing the linked maskal repository and the arXiv paper to understand the proposed active-learning software for Mask R-CNN in Detectron2. The issue names no Detectron2 files, tests, integration point, or acceptance criteria, so the intended change and definition of done are not established.

Written by the indexing model from the issue text.