The MS COCO Captioning Challenge is designed to spur the development of algorithms producing image captions that are informative and accurate. Teams were competing against each other by training their algorithms on the MS COCO 2014 dataset and their results will be scored by human judges.
Microsoft and Google’s research groups tied for first place in the recent MS COCO Image Captioning Challenge 2015. The winners were decided based on two main metrics: The share of captions that were equal to or better than a caption written by a person, and the share of captions that would pass a Turing test.
The Microsoft team outperformed competitors on the Turing test element, while the Google team won for the share of captions that were as good, or better, than what people could produce.
Read more about it here.