omniture

AInnovation Ranks First in Cityscapes, the World's Authoritative Instance Segmentation Challenge

2019-09-18 22:05 753

BEIJING, Sept. 18, 2019 /PRNewswire/ -- Recently, Cityscapes, the world's authoritative public evaluation datasets for instance segmentation, released its latest rankings. "AInnoSegmentation", the algorithm developed by AInnovation surpassed the performance of many well-known AI enterprises and university laboratories, including NVIDIA, Facebook, Uber, SenseTime, CUHK, Sogou and iFLYTEX.

AInnovation achieved three championships in 5 months, including WIDER FACE, PASCAL VOC and Cityscapes. It fully demonstrates the technological strength and algorithm innovation ability of innovative intelligence in the field of computer vision recognition.

Image from Cityscapes official website
Image from Cityscapes official website

About CITYSCAPES Datasets

Cityscapes datasets was provided by Mercedes-Benz in 2015, and is one of the most authoritative and professional instance segmentation datasets in computer vision. Cityscapes has two subtasks: pixel-level segmentation and instance segmentation, and instance-level segmentation is more challenging than pixel-level segmentation.

Cityscapes datasets contains a diverse set of stereo video sequences recorded in street scenes from 50 different cities, with high quality pixel-level annotations of 5,000 frames in addition to a larger set of 20,000 coarse annotated frames. The datasets have 30 classes, including ground, buildings, traffic signs, sky, people, vehicles and so on.

Up to now, Cityscapes has attracted nearly 100 teams, including Facebook, Uber, CUHK, SenseTime, Sogou and iFLYTEX. The AInnoSegmentation algorithm proposed by AInnovation has made breakthroughs in various indicators, and all the indicators are ranked first, and the comprehensive results are first.

About AInnoSegmentation Algorithm

AInnoSegmentation founded at famous Mask R-CNN network, using SE-Resnet 152 as backbone and six level FPN models to extract multi-level features. Then use the self-developed feature fusion modules as the feature fusion device, followed by two shared subnets, one for classification and bounding box regression, and one for image segmentation.

Mask R-CNN network
Mask R-CNN network

 

Performance of AInnoSegementation algorithm on Cityscapes dataset
Performance of AInnoSegementation algorithm on Cityscapes dataset

Commercial Value of Instance Segmentation

Since its founding, AInnovation has been aiming at the application of AI technology in the industries of manufacturing, retail and finance, focusing on developing advanced and mature AI algorithms, forming great commercial potential AI products and solutions.

The AInnoSegmentation algorithm is widely used in industrial vision and typical scenes including: complex scenes, such as defect detection, positioning, and recognition. For example, AInnovation has applied the algorithm to the quality inspection scenes such as garments and magnetic materials to improve the quality of the products.

The AInnoSegmentation algorithm can also be applied to product identification scenarios such as shelf intelligence and smart vending machine in the retail industry to improve the accuracy of product identification while helping customers improve operational efficiency.

In addition, the AInnovation also applies the AInnoSegmentation algorithm to the smart iron water unmanned locomotive transportation system of the steel plant, further improving the maturity and technical barriers of the solution.

About AInnovation and the Winner Team

AInnovation team who won the Cityscapes competition is composed of Faen Zhang, Jiahong Wu, Haotian Cao, Zhizheng Yang, Jianfei Song, Ze Huang, Jiashui Huang and Shenglan Ben. The team leader Faen Zhang is currently the CTO of AInnovation, the Chief Architect of AI Engineering Institute of Sinovation Ventures, and Honorary Professor of The University of Nottingham Ningbo China. He has worked at Microsoft, Google and Baidu, and made great achievements in the field of AI industry and academia. He holds many international AI patents and has published several top-level conference papers on AI. Other members of the team also have deep education background and practical experience with AI.

Established in March 2018, AInnovation is an AI subsidiary of Sinovation Ventures. With the mission of "AI Empowering Business", it is committed to providing AI-related products and business solutions for enterprises using the most advanced AI technology. Hocking Xu, CEO of AInnovation, has more than 20 years experience in sales, products, technology, services and management in the IT industry. He successively served as corporate executive to many of the world's top 500 technology giants including IBM, Microsoft and SAP, and has a deep understanding of the developments of various commercial fields in China. The AInnovation two-wheel model combines "Technology & Product" and "Business Scenarios", and gives rise to extremely fast commercialization.

There are many excellent R&D teams at AInnovation. Combined with the commercialization of AI technology, AInnovation has developed a three-level pyramid system for scientific and technical talents, including its Research Committee, its Research Institute and its Engineering and Algorithm Development Team, respectively, from top to bottom. Among them, "AInnovation Scientific Research Committee" is composed of the world's top artificial intelligence scientists and well-known experts, including Guangnan Ni, academicians of Chinese Academy of Engineering, Tong Zhang, former head of AI Lab of Tencent, Ruigang Yang, chief scientist of 3D vision of Baidu Research Institute, and Mi Zhang, professor of Michigan State University. Yonggang Wang, Executive Dean of the Innovation Workshop Artificial Intelligence Engineering Institute, and Faen Zhang, CTO of AInnovation.

Cision View original content to download multimedia:http://www.prnewswire.com/news-releases/ainnovation-ranks-first-in-cityscapes-the-worlds-authoritative-instance-segmentation-challenge-300920698.html

Source: AInnovation
collection