A team led by Associate Professor Yang Zhixin from the State Key Laboratory of the Internet of Things for Smart City and the Department of Electromechanical Engineering, Faculty of Science and Technology, University of Macau (UM), and Wang Shizheng, an associate professor from the School of Microelectronics, Chinese Academy of Sciences (CAS), recently won a second prize in the UG2+ Prize Challenge at the Institute of Electrical and Electronics Engineers (IEEE) Conference on Computer Vision and Pattern Recognition (CVPR 2019).
This year’s event attracted 260 research teams around the world, including artificial intelligence (AI) companies such as SenseTime, JingDong, Meitu Lab, and Hikivision. After the knockout stage, four research teams were left to compete for the final awards in a challenge titled ‘Object Classification on Video from Unconstrained Mobility Platforms’. The joint team formed by UM and CAS won the second prize of the competition for the performance of its research achievements based on the multidimensional test criteria. In addition, the team gave a speech titled ‘Selective Enhancement for Object Classification in Complicated Environment’. Members of UM in the team included PhD students Tang Lulu and Luo Luqing from UM’s State Key Laboratory of the Internet of Things for Smart City.
The UG2+ Prize Challenge was divided into two tracks, namely ‘Video Object Classification and Detection from Unconstrained Mobility Platforms’ and ‘Object Detection in Poor Visibility Environments’. Under the two tracks, there are five sub-challenges. The UM-CAS joint team joined a sub-challenge under the first track, titled ‘Object Classification Improvement on Video’. The challenge aimed to enhance the object classification accuracy on videos from unconstrained mobility platforms. The other awarded research teams included a team formed by the CAS and the Northeast University, as well as teams from Honeywell Intl, the Technical University of Munich, Sunway.AI: Joint Research Union, and Meitu Inc.
CVPR is one of the top three computer vision conferences in the world. The UG2+ Prize Challenge is co-organised by CVPR, related government departments, as well as academic and research companies in the field. The event aimed to evaluate and explore the accuracy and robustness of the most advanced computer vision algorithms for the tasks of object detection and classification in extreme conditions.
澳門大學智慧城市物聯網國家重點實驗室、科技學院機電工程系副教授楊志新與中國科學院微電子所副研究員王師崢帶領研究團隊，參與IEEE國際計算機視覺與模式識別會議（CVPR 2019）“UG2+計算機視覺算法挑戰賽＂，與來自世界各地科研機構共260多支參賽隊伍較量，最終脫穎而出，奪得全球亞軍。
今屆大賽吸引全球多家實力科研機構組隊參加，當中有來自商湯、京東、美圖和海康威視等國際著名的人工智慧企業。經過激烈的淘汰賽，最終僅剩四支隊伍角逐“無約束移動場景中的視頻目標分類”賽事的獎項。澳大中科院隊伍獲評審好評，其研究成果在多項不同維度的測試標準上表現甚佳，以第二名勝出。此外，團隊成員會上以“複雜環境下選擇性增強物體識別演算法”為題作主旨演說。是次參賽澳大成員包括澳大智慧城市物聯網國家重點實驗室博士生唐路路、羅鷺青。
是次挑戰賽共分“無約束移動場景中的視頻目標分類與檢測”和“可見性差場景中的目標檢測”兩大賽道，兩賽道下共設五個子賽道。其中，澳大中科院隊伍參加了前者賽道中的“視頻目標分類改進”子賽道，主要提升從無約束的移動場景中捕獲的視頻目標的分類準確率。今屆獲獎隊伍還包括中科院東北大學聯合隊（冠）、Honeywell Intl（季，去年冠軍）、慕尼克工大TUM、中科深微人工智慧實驗室、美圖實驗室Meitu Inc等。
CVPR為計算機視覺領域世界三大頂級會議之一；而“UG2+視覺算法挑戰賽＂由CVPR與計算機視覺領域知名的政府單位、學術機構和科技企業合辦，主要評估和探索當前最先進的影像處理演算法在極端條件下，對檢測和識別等任務的精確性與“魯棒性＂。