How to bring a deep learning inference to an embedded application?
The concept of pushing computing closer to where sensors gather data is a central point of modern embedded systems – i.e. the edge of the network. With deep learning, this concept becomes even more important to enable intelligence and
autonomy at the edge. For many applications – from automated machinery and industrial robots on a factory floor, to self-guided vacuums in the home, to an agricultural tractor in the field – the processing must happen locally.
The reasons for local processing can be quite varied depending on the application. Here are just a few of the concerns driving the need for local processing:
- Reliability. Relying on an internet connection is often not a viable option.
- Low latency. Many applications need an immediate response. An application may not be able to tolerate the time delay in sending data somewhere else for processing.
- Privacy. The data may be private and therefore should not be transmitted or stored externally.
- Bandwidth. Network bandwidth efficiency is often a key concern. Connecting to a server for every use case is not sustainable.
- Power. Power is always a priority for embedded systems. Moving data consumes power. The further the data needs to travel, the more energy needed.
TI addresses the need for bringing deep learning inference at the edge for embedded applications with a new reference design leveraging the highly integrated AM57x Sitara processors.
Deep Learning Inference for Embedded Applications Reference Design
This reference design demonstrates how to use TI’s deep learning library (TIDL-library) on a Sitara AM57x SoC to bring deep learning inference (or deep learning at the edge) to an embedded application. This design shows how to run deep learning inference on either C66x DSPs (available in all Sitara AM57x SoCs) and/or EVE processors, which are treated as black boxed deep learning accelerators on the Sitara AM5749 SoC.
• Automated sorting equipment
• Optical inspection
• Vision computer
• Code readers
• Industrial robots
• Logistics robots
• Currency counters
• Patient monitors
• Building automation
• Industrial transport
• Space, avionics & defense
Click here for the reference design.