Summary of KONTRON AI PLATFORM BASED ON GOOGLE CORAL EDGE TPU FOR ARTIFICIAL INTELLIGENCE
Kontron’s AI platform combines the NXP i.MX8M and Google Coral Edge TPU to run TensorFlow Lite applications, enabling fast development of ML and DL models with optimized time-to-market. The Coral Edge TPU delivers 4 TOPS, accelerating image/video tasks (e.g., from ~6 fps to ~30 fps). The system uses an open-source Yocto-based Linux image (Kernel 4.18) on eMMC, supports Python and C++, and leverages model quantization (32-bit to 8-bit) for efficient Edge TPU execution and low-latency edge computing.
Parts used in the Kontron AI platform:
- NXP i.MX8M processor
- Google Coral Edge TPU
- TensorFlow Lite framework
- Pre-trained TensorFlow Lite models
- Yocto-based Linux image (custom, Kernel 4.18) on eMMC
- Python and C++ programming environments
- USB cameras (example application without TPU)
- Quantized 8-bit model representations
Kontron AI platform based on NXP i.MX8 M & Google Coral Edge TPU allows easy integration of all Google TensorFlow-Lite applications for Artificial Intelligence. The applications of TensorFlow Lite and pre-trained models can be downloaded for free of cost. It also comes with the feature of quick simple development of neural networks with machine learning and deep learning applications and thus has an optimized TTM (Time To Market).
The Google Coral Edge TPU (Tensor Processing Unit) enhances the AI platform, performing 4 TOPS (Trillion Operations Per Second) for high-speed image and video data processing. This enables the development of industrial AI applications such as object recognition and classification and other image processing-based applications.
Compared to an application with simple USB-cameras without TPU at approx. 6 frames/s, the TPU accelerates to a speed of 30 frames/s, therefore five times faster.
The software is an open-source custom Linux-based system with Yocto based Linux image in eMMC Kernel 4.18. Kontron AI platform is compatible with the TensorFlow Lite framework with Python and C++ as the primary programming languages. Models of the framework feature compactness for more efficiency, through quantization. This converts 32-bit parameter data into 8-bit representations which are suitable for TPU. This enables Edge Computing for low latency and improves the performance significantly.
Read more: KONTRON AI PLATFORM BASED ON GOOGLE CORAL EDGE TPU FOR ARTIFICIAL INTELLIGENCE
- What processors does the Kontron AI platform use?
It uses the NXP i.MX8M processor together with the Google Coral Edge TPU. - Can I run TensorFlow Lite applications on this platform?
Yes, the platform is compatible with the TensorFlow Lite framework. - What operating system does the platform use?
It uses an open-source custom Yocto-based Linux image with Kernel 4.18 on eMMC. - Which programming languages are supported?
Python and C++ are the primary programming languages supported. - How much performance does the Edge TPU provide?
The Google Coral Edge TPU delivers 4 TOPS for high-speed processing. - Does the TPU improve camera frame rates?
Yes, an application that ran at approximately 6 frames per second without TPU can reach about 30 frames per second with the TPU. - How are models optimized for the TPU?
Models are quantized from 32-bit to 8-bit representations to be suitable for the TPU and more efficient. - Are TensorFlow Lite models available cost-free?
Yes, TensorFlow Lite applications and pre-trained models can be downloaded free of cost.
