DRP-AI TVM Makes It Easy to Port AI Model into RZ/V Devices
Have you ever found it difficult to implement an AI model in an embedded device? We would like to introduce DRP-AI TVM[1] for easy porting of AI models.
In recent years, the implementation of AI in embedded devices such as surveillance cameras, traffic monitoring systems, robots, and drones has been increasing. Unlike the cloud environment, embedded devices have many considerations such as HW performance and restrictions in cost and power consumption, making AI implementation difficult. By using the AI-MPU RZ/V series, which implements Renesas' low-power AI accelerator DRP-AI, most of the hardware implementation issues can be resolved.
When users implement AI models into MPUs, they can use the DRP-AI translator to convert AI models into DRP-AI executable format. However, we received feedback from some users about the difficulty of implementation.
The main reasons are two：
AI model format supported by the DRP-AI translator is ONNX only.
If non-supported AI model operators are included in AI models, the DRP-AI translator cannot convert it into an executable format.
Renesas has asked users to handle such cases themselves, which has caused a burden on some users. We released DRP-AI TVM to solve the issues and to improve user experience.
DRP-AI TVM is the tool that added DRP-AI support on an open-source ML compiler Apache TVM. Using DRP-AI TVM, AI model operators non-supported by DRP-AI translator can be automatically allocated to CPUs to implement more AI models into RZ/V products easily. DRP-AI TVM supports not only ONNX, and other AI frameworks such as Poarch and TensorFlow, etc.
Furthermore, as with the conventional DRP-AI translator flow, DRP-AI can handle the pre-processing of AI inference, enabling total AI performance improvement, including pre-processing.
DRP-AI TVM structure
Auto AI model operators’ allocation by DRP-AI TVM
DRP-AI TVM is public on GitHub for users to ease to access. Users can find sufficient information to use DRP-AI TVM on GitHub, such as the tutorial, sample code, and a variety of AI model performances. The tutorial includes steps from building the environment to AI inference on an evaluation board, and customers can use these steps to evaluate their AI models easily. Renesas will continue to strive for UX improvements to make it easier for customers to use Renesas products.
- +1 Like
- Add to Favorites
Recommend
- Renesas Unveils RZ/V Microprocessor Series with Vision-Optimized Artificial Intelligence Accelerator
- Renesas‘ Entry-Level RZ/V2L MPUs with Best-in-Class Power Efficiency and High-Precision AI Accelerator
- Renesas DSP Solution on Renesas Lab on the Cloud, Input Analog Signals Directly From A Signal Generator To The Rx231 Microcontroller Evaluation Board
- Renesas RZ/V2M with DRP-AI Wins 2020 Aspencore‘s World Electronic Achievement Award (WEAA)
- Renesas‘ Semiconductor Manufacturing Factory (Naka Factory) Fire: Production Capacity Has Recovered to 88%
- Check Out the New Features in Renesas RZ/V MPU AI Applications v2.00!
- Renesas Announces 10 New Winning Combinations Integrating Celeno and Renesas Products
- Renesas ＆ Altran to Deploy First Social Distancing Wristwatch Using Ultra-Wideband Chipset with Low Rate Pulse
This document is provided by Sekorm Platform for VIP exclusive service. The copyright is owned by Sekorm. Without authorization, any medias, websites or individual are not allowed to reprint. When authorizing the reprint, the link of www.sekorm.com must be indicated.
Integrated Circuits
Discrete Components
Connectors & Structural Components
Assembly UnitModules & Accessories
Power Supplies & Power Modules
Electronic Materials
Instrumentation & Test Kit
Electrical Tools & Materials
Mechatronics
Processing & Customization
