China's AI-focused company sets new world records

China's AI-focused company Cloudwalk set new world records across four datasets, PA-100K, RAP V2, PETA, and HICO-DET, with its foundation model for pedestrian analysis, marking the technology reaching the standard of large-scale commercialization for the first time.

The Chongqing entity of Cloudwalk is located at the Liangjiang Digital Economy Industrial Park. [Photo/Feng Zhiyi]

The pedestrian foundation model is a cornerstone in the realm of visual foundation models, catalyzing the development of sophisticated foundation models with a focus on computer vision. These models have made strides in pedestrian attribution recognition, encompassing a range of features including holistic traits such as gender and age as well as partial features such as clothing and accessories, carried-on items like mobile phones, knives, and handbags, and nuanced human-object interactions such as smoking, holding a knife, and taking a photo. An unprecedented four record-breaking datasets have been curated to comprehensively cover these domains, thereby facilitating a more comprehensive understanding of pedestrian behavior and improving the efficacy of computer vision systems.

The traditional technology of pedestrian detection and analysis has disadvantages, including the need for more understanding of the scenario and can only detect all targets. Secondly, subtle features of pedestrians require fine-grained recognition ability and are easily affected by large gestures and environmental disturbances; last, with varied demands of pedestrian detection and analysis, customized developments are needed more often than not. Consequently, reducing the development cost becomes critical for the technology to industrialize.

The AI large model independently developed by Cloudwalk enables a significant decrease in data dependency and increases in developers' training efficiency. It can achieve a similar result with the customized model of the original scenarios by using 1 percent of its data. It is suitable for industries where the costs for actual samples are fairly high. The technology has been applied for safety purposes in various scenarios, such as mines and construction sites.

Cloudwalk provides man-machine coordination solutions based on integrated industry-leading AI, cognitive computing, and big data technology. As the largest AI supplier in China's financial industry, it has explored four major business areas, including AI finance, AI governance, AI transportation, and AI business.

Earlier this year, the company secured significant progress in its strategic layout of "big data, strong computing power, and powerful algorithms" by unveiling the western center of smart computing in Shuitu New City in the Liangjiang New Area. The center is set to operate the production line of foundation models. The company plans to launch industry-oriented foundation models and provide end-to-end solutions to cover various areas, including finance, government, travel, energy, education, medical, and cultural entertainment.