Mask-RCNN is an image segmentation model that has been trained on the COCO dataset. It not only detects objects in image, but also works out which pixels belong to that object to create a mask. I've implemented this in PyTorch and added in a feature to grey out the background.
