Summary of TIDEP-01004 – MACHINE LEARNING INFERENCE FOR EMBEDDED APPLICATIONS REFERENCE DESIGN
This reference design shows how to run TI Deep Learning (TIDL) inference on Sitara AM57x SoCs, using C66x DSP cores and EVE subsystems (including AM5749). It provides optimized CNN models, a full TIDL development flow (training, import, deployment), benchmarks, and a getting-started guide in the AM57x Processor SDK for embedded deep learning inference and performance evaluation.
Parts used in the Machine Learning Inference for Embedded Applications Reference Design:
- Sitara AM57x System-on-Chip (SoC)
- AM5749 SoC (example target)
- C66x DSP cores (on AM57x)
- Embedded Vision Engine (EVE) subsystems (on AM5749)
- TIDL (TI Deep Learning) library
- Reference CNN models for classification, detection, segmentation
- AM5749 IDK EVM (evaluation module)
- AM57x Processor SDK (software/guide)
This reference design demonstrates how to use TI Deep Learning (TIDL)/Machine Learning on a Sitara AM57x System-on-Chip (SoC) to bring deep learning inference to an embedded application. This design shows how to run deep learning inference on either C66x DSP cores (available in all AM57x SoCs) and Embedded Vision Engine (EVE) subsystems, which are treated as black boxed deep learning accelerators on the AM5749 SoC.
This reference design is applicable to any application that is looking to bring deep learning/machine learning inference into an embedded application. Customers looking to quickly get started with a deep learning network or to evaluate their own networks performance on an AM57x device will find a step-by-step guide on how to use TIDL available as part of TI’s free AM57x Processor SDK.
Features
- Embedded deep learning inference on AM57x SoC
- Performance scalable TI deep learning library (TIDL library) on AM57x using C66x cores only, EVE subsystems only, or C66x + EVE combination
- Performance optimized reference CNN models for object classification, detection and pixel-level semantic segmentation
- Full walk-through of TIDL development flow: training, import and deployment
- Benchmarks of several popular deep learning networks on AM5749
- This reference design is tested on AM5749 IDK EVM and includes TIDL library on C66x core and EVE subsystem, reference CNN models and getting started guide
Read more: TIDEP-01004 – MACHINE LEARNING INFERENCE FOR EMBEDDED APPLICATIONS REFERENCE DESIGN
- What does this reference design demonstrate?
It demonstrates using TI Deep Learning (TIDL) on Sitara AM57x SoCs to run deep learning inference on C66x DSP cores and EVE subsystems. - Which processors or accelerators are targeted?
It targets C66x DSP cores available on all AM57x SoCs and EVE subsystems on AM5749. - Is this reference design applicable to other embedded applications?
Yes, it is applicable to any application seeking to add deep learning inference to an embedded system. - What components of TIDL usage does the design cover?
It covers the full TIDL development flow: training, import, and deployment. - Does the design include example CNN models?
Yes, it includes performance-optimized reference CNN models for classification, detection, and pixel-level semantic segmentation. - Are performance benchmarks provided?
Yes, the design includes benchmarks of several popular deep learning networks on AM5749. - What hardware was the reference design tested on?
It was tested on the AM5749 IDK EVM. - Where can users find a step-by-step guide?
The step-by-step guide on using TIDL is available as part of TI's free AM57x Processor SDK.
