Neural Networks are getting better and better. Recently Facebook has published “detectron” – an open source software to detect objects and draw segmenation-masks or keypoint annotations around them. Detectron uses a Mask R-CNN architecture and is trained on the MS COCO dataset. It is quite easy to use – you can just start a docker container and you have your Caffe2 and python environment set up and running.
To show how well detectron is working I have created a small demo. We have a drone flying through our HQ and apply detectron on the video. See the results here: