31
2022
-
03
ICCV 2021 | DeepBlue's Computer Vision Technology Claimed Another 3 World-class Titles
From October 11th to October 17th, the 2021 International Conference on Computer Vision (ICCV), known as one of the three top conferences in the global computer vision field, was held as scheduled. The DeepBlueAI team of DeepBlue Technology participated in 2 competitions and 4 challenges and won 1st place in the VisDrone Object Detection, VisDroneMot, Large-AI-Food, and VisDrone. Among them, VisDrone has become a benchmark dataset in the field of drones. And many papers in the industry have also been published based on this dataset.
Views:
From October 11th to October 17th, the 2021 International Conference on Computer Vision (ICCV), known as one of the three top conferences in the global computer vision field, was held as scheduled. The DeepBlueAI team of DeepBlue Technology participated in 2 competitions and 4 challenges and won 1st place in the VisDrone Object Detection, VisDroneMot, Large-AI-Food, and VisDrone. Among them, VisDrone has become a benchmark dataset in the field of drones. And many papers in the industry have also been published based on this dataset.
The two competitions attracted many well-known teams from around the world, including many major universities and top technical teams, including Tsinghua University, Institute of Computing Technology, Chinese Academy of Sciences, Beijing University of Posts and Telecommunications, University of Barcelona, Tencent, Google, Alibaba, OPPO, and hundreds of other well-known technical teams.
It is worth mentioning that, as a "regular customer" of ICCV, the DeepBlueAI team of DeepBlue Technology has won many other top international competitions with its leading technical level. Won dozens of championships in ACL, NAACL, and other competitions.
1st Place in Drones Competition
VisDrone Becomes the Benchmark Dataset in the Field of Drones
In the VisDrone Object Detection competition project, there are two challenges, "target detection in images" and "multi-target tracking challenge". The goal is to detect objects of predefined classes (e.g., cars and pedestrians); while the task of the Multi-Object Tracking Challenge aims to recover the trajectories of objects in each video frame.
VisDrone Object Detection (Winner: DeepBlue)
VisDroneMot Challenge (Winner: DeepBlue)
The VisDrone dataset was collected by the AISKYEYE team of the Machine Learning and Data Mining Laboratory of Tianjin University. All benchmark datasets were captured by drones, including 288 video clips, with a total of 261,908 frames and 10,209 still images. The frames consist of more than 2.6 million manually annotated boxes for common objects such as pedestrians, cars, bicycles, and tricycles.
Although the competition has been held many times, the difficulties still persist:
1. Numerous detected objects
2. Some targets are too small
3. Different data distributions
4. The target is severely occluded
Finally, the DeepBlue Technology team completed the project competition by adding noise randomly, changing the brightness, using center cropping, mosaic data enhancement, along with other methods, and won 1st place on these two challenges.
1st Place in Large-scale Fine-grained Image Retrieval
In addition, in the LargeFineFoodAI technical seminar co-hosted by Meituan Visual Intelligence Center, Institute of Computing Technology of the Chinese Academy of Sciences, Beijing Zhiyuan and University of Barcelona, the first LargeFineFoodAI competition also kicked off. The competition is divided into two challenges, Recognition and Retrieval. It aims to use computer vision algorithms to conduct fine-grained analysis of food images to quickly respond to and meet the needs of merchants and users for reviewing, managing, browsing, and evaluating online food images.
Large-Scale Fine-Grained Food Retrieval Challenge (Winner: DeepBlue)
The data set used in this challenge comes from Meituan's self-built data set "Food2K". Each food picture in this data set is taken by different individuals, using different equipment, and shooting in different environmental scenarios. It is a rare algorithm that can be fairly evaluated. The robustness and effect of picture data pose great challenges to the participants.
DeepBlueAI participated in the competition and used the ReRank method to weight the Euclidean distance and the Jacobian distance to measure the similarity between the query and the gallery, and achieved good results in the Large-ScaleFine-Grained Food Retrieval track. Among them, China Science The technical university and the OPPO company team won second and third place respectively.
ICCV, also known as IEEE International Conference on Computer Vision (International Conference on Computer Vision) is hosted by IEEE and is held every two years around the world. Together with the Conference on Computer Vision Pattern Recognition (CVPR) and the European Conference on Computer Vision (ECCV), it is widely known to be the three top conferences in the direction of computer vision.
DeepBlue Technology has been in ICCV for many years and has achieved remarkable results. When the global economy has entered the era of the fourth industrial revolution led by AI and China is focusing on new AI infrastructure, DeepBlue Technology will always be ready to apply core technologies such as computer vision in basic research and product development and use technology to make people's lives better.
Key words:
Related News
2022-10-27
2022-10-27
2022-10-27











LinkedIn Page
Copyright © 2022 DeepBlue Technology (Shanghai) Co., Ltd