On the morning of October 29th local time, Joint COCO and Places Recognition was held at the International Conference on Computer Vision (ICCV 2017) in Venice, Italy The COCO and Places rankings were announced at the Challenge “Workshop. Megvii participated in the most important four of the seven challenges and won three first prizes and one second prize, beating companies from Microsoft, Facebook, Google, Sensetime, etc. And Carnegie Mellon University, Peking University, the Chinese University of Hong Kong, Shanghai Jiao Tong University and other universities, became the first Chinese company to win the COCO competition.

COCO Challenges Ranking Result

The MS COCO (Microsoft Common Objects in Context) Challenge is one of the most closely watched and authoritative competitions in computer vision since ImageNet. It is the most important benchmark for image (object) recognition, and the only competition in the international field that can bring together Google, Microsoft, Facebook, top international universities and excellent innovative enterprises. Compared with the ImageNet image classification task that focuses on the whole image, the object detection task in COCO pays more attention to the individual of each object in the image (such as various small objects and various occlusion objects), so the algorithm is required to have a better understanding of the image details. This competition also represents the highest level of image recognition since ImageNet.

Megvii Face++ team

COCO has been held three times since 2015, and the first two object detection champions are MSRA and Google respectively. This year’s COCO contains four tasks: Detection Challenge (object Detection), Instances segmentation (object segmentation), (Human) Keypoint Challenge (Human Keypoint Detection), Stuff segmentation (background semantic segmentation). Megvii Face++ team participated in the challenges of the first three tasks and achieved two first (object detection, human key points) and one second (object segmentation).

Places, an international computer vision competition for deep understanding of graphic scenes, was launched this year by MIT and CMU, in collaboration with COCO. There are three tasks in Places 2017: Scene Parsing, Instance Segmentation, and Semantic Boundary Detection. The megvii Face++ team only took part in the object segmentation challenge, beating arch-rival Google to win the task.

For industry, machines are important for understanding people, objects and scenes. Megvii’s achievements in the COCO and Places competitions attest to the global leadership of Megvii. Using technical competitive advantage, kuang inspect institute jointly with the product center will continue in the product development to promote technology transfer, behavior recognition and scene segmentation, object detection and segmentation technology in the Internet finance, intelligent security, cities such as brain, new retail, mobile phones, the application of practical scenarios or industry such as exploration, in order to realize the technology value maximization.

Sun Jian, chief scientist and director of the research Institute of Megvii, said that megvii was able to win the three most important titles on behalf of a Chinese enterprise in the most competitive competition for the first time, mainly relying on three magic weapons:

  1. MegBrain, a deep learning engine developed by Megvii and used by all employees, enables us to systematize training algorithms at the fastest speed.

  2. Megvii research Institute has accumulated long-term and in-depth research on deep learning and computer vision algorithms. Although Megvii rarely “tops the list” in international data set competitions, in fact, the internal technical indicators are always very high, so I take this opportunity to share them with you.

  3. In addition to abundant computing resources, megvii research Institute has an environment that encourages continuous high-speed innovation and a culture of pursuing perfection.

“Finally, I wish our team in Venice, I am proud of you, proud of Chinese technology enterprises!”

Power  Human  with  AI.

Long press the qr code to follow kuang vision (Face++)

Let the machine understand the world

The world’s leading image recognition service

www.megvii.com

✆ 400-6700-866