‘Democratizing deep learning – a low cost food product identification approach designed for embedded devices’

The EMVA is happy to welcome Dr. Vassilis Tsagaris, CEO at IRIDA Labs, presenting a use case from the food industry at Embedded VISION Europe conference.

Vassilis Tsagaris founded IRIDA Labs in 2009 with the aim to bridge the gap between a camera and a human eye by bringing visual perception to any device, making embedded computer vision accessible to everyone. Today, as CEO of IRIDA Labs he is responsible for corporate strategy, business partnerships and business development worldwide. IRIDA Labs is partnered with large international companies like Qualcomm, Cadence and CEVA (all NASDAQ listed) and has a client basis in Europe, China and USA.
Prior founding IRIDA Labs, he has worked as a researcher, post-doc researcher or project manager for about ten European and National R&D projects for the academic and company sector. He has BSc. In Physics and a PhD in Computer Vision and Data Fusion from the University of Patras, Greece, and has published more than 30 journal and conference papers

Abstract of  Vassilis’ presentation:

Deep learning has recently emerged as the dominant approach for performing various classification tasks ranging from computer vision to speech processing. For computer vision, Deep Convolutional Neural Networks (CNNs), are incorporating end-to-end learnable modules able to achieve robust feature representations. However, CNN based approaches developed by technology giants like Google, Baidu or others often require large amounts of data for training and are computationally intensive during evaluation, which makes them impractical or even prohibitive for embedded or time-critical applications.
In this presentation we are going to present how we democratize deep learning by studying the application scenario and hardware platform in order to be able to transfer the knowledge and accuracy of large scale CNN networks in an embedded device, thus making deep learning a powerful tool for everybody.
A case study is presented for the food recognition scenario where we have conducted analysis utilizing the FOOD 101 database which is comprised by images of food taken into different conditions and it is organized to 101 categories. We will first present the results of our CNN-based approach outperforming conventional approaches and then we will discuss how we implement the inference or evaluation part of the CNN structure in a common ARM based CPU embedded system achieving the low power and high speed performance needed for this case study.
Finally, we will discuss how this approach is applied in a food preparation environment in order to categorize between predefined products (like bakery products) in an unconstrained environment where embedded deep learning at the edge provides breakthrough solutions to challenging computer vision problems.

The debut of EMVA’s brandnew conference Embedded VISION Europe, supplemented by an already well booked table top exhibition, will take place 12-13 October 2017 in Stuttgart.

Find all conference details at www.embedded-vision-emva.org