Deep Learning platform speeds projects with Renesas R-Car SoCs
Chipmaker Renesas collaborates with Fixstars Corporation, a provider of multicore CPU/GPU/FPGA acceleration technology, in the field of Deep Learning in the automotive sector. The two companies plan to launch an Automotive SW Platform Lab in April 2022 to develop software and operating environments for Renesas automotive SoCs. The Lab will support the entry into development and advancement of advanced driver assistance systems (ADAS) and autonomous driving (AD) systems. The partners will jointly develop technologies for Deep Learning software development and building operating environments, Renesas said. This will allow learned network models to be continuously updated to ensure and improve recognition accuracy and performance.
Fixstars has both state-of-the-art Deep Learning software technology and optimisation technology that enables more efficient use of hardware, according to Renesas. “Renesas customers can now take full advantage of the superior performance of Renesas’ automotive SoCs,” explained Takeshi Kataoka, general manager of the Automotive Solution Business Unit at Renesas
“After developing a deep-learning application, it is not possible to maintain high recognition accuracy and performance without constantly updating it with the latest machine learning data,” said Satoshi Miki, CEO of Fixstars. “Fixstars is focusing on these automotive machine learning operations (MLOps) in collaboration with Renesas to develop a deep learning development platform optimised specifically for Renesas SoCs.”
As part of their collaboration, Renesas and Fixstars are introducing “Genesis for R-Car”, a cloud-based evaluation environment for R-Car V3H SoCs. It supports entry-level ADAS and AD system development. The new environment enables immediate initial evaluation when selecting SoCs. It uses Fixstars’ cloud-based evaluation environment Genesis as a platform.
The background: Reviewing specifications by human experts is time-consuming and inefficient. However, when selecting SoCs, evaluation based on real use cases is essential. Usually, users need to purchase an evaluation board and basic software to evaluate SoCs. In addition, technical expertise is also required to set up an evaluation environment. The new cloud-based evaluation environment does not require any specific technical expertise.
With Genesis for R-Car SoCs, developers can check the processing time in frames per second (fps) and the percentage recognition accuracy of the CNN accelerators from the R-Car V3H SoC using sample images with generic CNN models, such as ResNet or MobileNet. Developers can also select the SoC and network they want to evaluate and perform operations remotely on a real board. The Genesis environment can be used to validate evaluation results in image classification and object recognition, for example. Developers can also use their own images or video data. This greatly simplifies the initial evaluation of how suitable R-Car V3H SoCs are for the customer’s system. In the future, it is planned to introduce a service that will allow users to use their own CNN models for evaluations.