Shulou(Shulou.com)11/24 Report--

Recently, sponsored by China Society of Image Graphics (CSIG) and organized by Tianyi Cloud Technology Co., Ltd., the 4th CSIG Image Graphics Technology Challenge "Pedestrian Attribute Recognition" track announced the results of the competition. Tianyi Cloud algorithm engineer Lu Jiada presented awards to the winners. The challenge aims to encourage researchers and technical teams to carry out innovative research in the field of pedestrian attribute recognition through competition, so as to further promote the development and application of relevant technologies.

Pedestrian attribute recognition refers to the process of judging which attributes pedestrians possess based on pedestrian images. It can identify and track pedestrian characteristics in images or videos, such as gender, age, color and style of clothing, and whether there are accessories such as knapsacks and hats. It is widely used in dangerous behavior warning, traffic violation monitoring, industrial security, and target locking in public places. The application of pedestrian attribute recognition can greatly improve the efficiency of extracting effective information from surveillance video, reduce the consumption of manpower and material resources, and play an indispensable role in building a safe city, maintaining public safety and improving the level of social governance.

The "Pedestrian Attribute Recognition" Track Challenge organized by Tianyi Cloud was open to the whole society, attracting 97 teams and five teams to participate in the competition. Tianyi Cloud provided the contestants with a complete set of labeled pedestrian attribute images, each of which was labeled with 15 attributes, including dress style, age, gender, orientation, etc. Competitors are required to design algorithmic models based on this dataset to identify pedestrian attributes in the dataset, which requires contestants to build a sufficiently robust multitask model to complete 15 tasks simultaneously and deal with sample imbalances between different tasks. In this challenge, the contestants demonstrated their innovative achievements and excellent ability in the field of pedestrian attribute recognition through wonderful works and projects, which not only trained the contestants 'ability to solve real scene problems, but also provided new solutions and technical methods for the field of fine-grained image classification.

After fierce competition, three teams stood out and won the championship, second place and third place respectively. Among them, EricTan team from University of Chinese Academy of Sciences won the championship with F1 score of 0.815, and participated in the finals as the representative of this stage and other track champions. Different from the traditional pedestrian attribute recognition scheme, the scheme adopted by this champion is to generate text description for each attribute label by using powerful GPT-4 model, and learn corresponding text features by using Transformer encoder, and calculate the correlation with the features of pictures. By exploring the correlation between attribute semantics and image space, this scheme achieves better results than other attribute recognition algorithms, and provides a new working idea for other similar downstream tasks.

As the origin of cloud computing original technology, Tianyi Cloud always insists on scientific and technological innovation and independent control of key core technologies. With powerful cloud computing and data processing capabilities, as well as rich technical resources accumulated in image processing and other fields, Tianyi Cloud provides stable and reliable computing platform and data support for pedestrian attribute recognition research and application. In the past three years, Tianyi Cloud has submitted a number of patents in the fields related to pedestrian attribute recognition, with profound technical accumulation. At present, Tianyi Cloud pedestrian attribute algorithm has been implemented, and has been applied and tested in the company and customers, and has received unanimous praise.

In the future, Tianyi Cloud will continue to adhere to key core technologies independently, further promote the development and application of pedestrian attribute recognition technology, assist pedestrian attribute recognition to play a role in a wider range of scenarios such as public safety and commercial fields, and contribute more to the construction of smart cities and digital China.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.