NeurIPS 2019
Sun Dec 8th through Sat the 14th, 2019 at Vancouver Convention Center
Paper ID:3565
Title:Weakly Supervised Instance Segmentation using the Bounding Box Tightness Prior

Although the bounding box supervision is relatively more costly than the simple image-level class labels, the proposed method exploits the benefit of the given annotation effectively and achieves good performance. In particular, the MIL formulation is interesting. I recommend accepting this paper.