World's best Coco

The Microsoft COCO Captioning Challenge is designed to spur the development of algorithms producing image captions that are informative and accurate. There are 18 teams all together and our attributes based image captioning framework currently achieves the best result on 3 evaluation metrics (BLEU-1,2,3) out of 7. We also achieve the top-5 ranking on the other evaluation metrics. See https://www.codalab.org/competitions/3221#results for more information. Our algorithm is labelled as Q.Wu.

In the following examples our generated captions are in black. The baseline is in blue, and a human caption is in red. Attributes and corresponding scores are shown at the side.

Coco
Tagged in Award, Challenge, Research, Machine Learning, Computer Vision