Is CenterNet possible applied for segmentation?

[NeuZhangQiang]

CenterNet is really good network, and this project simply show me how it work.

Now, I want to apply CenterNet for image segmentation. Just for Mask-RCNN, it can simultaneously perform object detection and semantic segmentation. I want to add a branch in CenterNet to perform the segmentation. But, I don't know how to start it, could you please provide some suggestions?

Currently, the CenterNet has three branches: heatmap (128*128*C), offset (128*128*2) and size (128*128*2). How to design the new segmentation branch? Is it possible to generate a seg_prob (128*128*(128*128))? It indicates to predicted a mask (128*128) for each point in the grid (128*128).