Closed munsanje closed 5 years ago
I haven't looked at it extensively, but this was implemented without that paper. The concept of this repository is that it follows MaskRCNN principles, but uses RetinaNet as an object detection backbone as opposed to Faster-RCNN. It seems to resemble RetinaMask a lot, but I guess there are bound to be subtle differences.
Actually looking at their release dates and our history, it is more likely that they researched our work ;p
It doesn't say in the documentation, but this seems to be an implementation of Fu et al's RetinaMask, implemented here and described here. Am I correct?