From March 31st to April 2nd, the China Electric Vehicle 100 Forum (2023) hosted by the China Electric Vehicle 100 was held in Beijing. With the theme of “promoting the modernization of China’s auto industry”, this forum invites representatives from all walks of life in the fields of automobile, energy, transportation, city, communication, etc. Discussions will be held on many cutting-edge topics in the automotive industry, such as trends and high-quality development paths for new energy vehicles.
As a representative of the cloud computing field, You Peng, Director of EI Service Product Department of Huawei Cloud Computing Company, was invited to give a keynote speech at the Smart Car Forum. He said that there are many business pain points in the development of business requirements in the field of autonomous driving, and creating a closed loop of autonomous driving data is the only way to achieve high-level autonomous driving. HUAWEI CLOUD provides a three-layer acceleration solution of “training acceleration, data acceleration, and computing power acceleration” to enable efficient training and inference of models, and realize fast closed-loop circulation of autonomous driving data.
You Peng said that with the continuous accumulation of intelligent driving mileage, the generation of massive driving data means that the level of intelligent driving will develop higher. But at the same time, the challenges faced by autonomous driving companies are becoming increasingly apparent. Among them, how to manage massive data, whether the tool chain is complete, how to solve the problems of computing resource shortage and conflict with computing power, and how to achieve end-to-end security compliance have become the pain points that need to be faced in the development process of autonomous driving. question.
You Peng mentioned that among the key factors affecting the implementation of autonomous driving at present, there are “long tail problems” in various uncommon but emerging scenarios. Therefore, large-scale and efficient processing of new scenario data and rapid optimization of algorithm models have become automatic The key to iteration of driving technology. HUAWEI CLOUD provides three-layer acceleration of “training acceleration, data acceleration, and computing power acceleration” for the pain points in the autonomous driving industry, which is an effective solution to the long-tail problem.
1. The “ModelArts Platform” that provides training acceleration can provide the industry’s most cost-effective AI computing power. HUAWEI CLOUD ModelArts’ data loading acceleration DataTurbo can implement reading while training, avoiding bandwidth bottlenecks between computing and storage; in terms of training and inference optimization, model training acceleration TrainTurbo automatically integrates trivial operator calculations based on compilation optimization technology, which can achieve One line of code optimizes model calculations. With the same computing power, efficient training and reasoning can be achieved through the ModelArts platform.
2. Provides large model technology as well as NeRF technology for data generation. Data labeling is a relatively expensive link in the development of autonomous driving. The accuracy and efficiency of data annotation directly affect the efficiency of the algorithm. The large-scale labeling model developed by Huawei Cloud is pre-trained based on massive typical data. Through semantic segmentation and object tracking technologies, it can quickly complete the automatic labeling of long-term continuous frames and support subsequent automatic driving algorithm training. The simulation link is also a link with a high cost of autonomous driving. Huawei Cloud NeRF technology greatly improves the efficiency of simulation data generation and reduces simulation costs. This technology ranks first in the international authoritative list, and has obvious advantages in image PSNR and rendering speed.
3.HUAWEI CLOUD Ascend cloud service that provides computing power acceleration. Ascend cloud service can provide safe, stable and cost-effective computing support for the autonomous driving industry. Ascend Cloud supports mainstream AI frameworks, and has made targeted optimizations for typical models of autonomous driving. The convenient conversion toolkit facilitates customers to quickly complete the migration.
In addition, HUAWEI CLOUD relies on the “1+3+M+N” global automotive industry cloud infrastructure layout, that is, a global automotive storage and computing network, 3 super-large data centers to build a dedicated automotive area, M distributed IoV nodes, N A car-specific data access point, helping enterprises build data transmission, storage, computing, professional compliance infrastructure, and helping the car business to go global.
HUAWEI CLOUD will continue to practice the concept of “everything is a service”, adhere to technological innovation, provide more complete solutions for the autonomous driving industry, and work with partners to provide customers with cloud empowerment, and continue to contribute to the innovation and development of global autonomous driving.
Post time: Apr-03-2023
