How to run a CNN TensorFlow model on an embedded hardware accelerator
How to run a CNN TensorFlow model on an embedded hardware accelerator

How to run a CNN TensorFlow model on an embedded hardware accelerator

Summary:

In this video we will show a celebrity comparison demo and explain how to train such a model. We will look into embeddings and how to use a deep neural network to create bottleneck features resulting in image comparison.  

Further, we will explain how to preprocess the model to be ready for the neural accelerator on an embedded hardware (phyBOARD pollux, NXP i.mx8M Plus).

Published By
Jan Werth (PhD)
Jan Werth was born and raised in Mainz am Rhein. He moved to Stuttgart in 2003 to become a professional cabinetmaker. Four years later, he entered his studies for electrical engineering with the focus on signal processing and systems in the University of Applied Sciences in Trier. He finished his master in Eindhoven with Philips Research on the topic of automated pneumonia detection following up with a PhD with the Eindhoven University of Technology, the Maxima Medical Center, and Philips Research. Here he focused on automated preterm infant sleep analysis. Currently, he works for Phytec Messtechnik GmbH in Mainz as lead data scientist, where he is responsible for all data driven solutions. Jan Werth has over ten years of experience in signal processing, with a focus on machine- and deep learning since 2013. Detailed information can be found under linkedin.com/in/jan-werth... Show more
Comment