Data: mercoledì 11 ottobre 2017
Orario: dalle 8:30 alle 17:30
Luogo: Mappa
Costo: Gratuito


T3LAB partecipa all’evento con un dimostratore all’interno della sessione pomeridiana.
Nel contesto di questo evento T3LAB presenterà uno dei dimostratori realizzati come parte del progetto di ricerca finanziata Open-Next (fondi POR-FESR).
Il dimostratore illustra le potenzialità di un sistema per l’esecuzione accelerata su FPGA (in particolare su SOC come Zynq) di reti neurali convoluzionali (CNN) progettate e trainate utilizzando i framework di machine learning convenzionali, come Caffe e TensorFlow.
Il sistema è composto di due parti: l’acceleratore Nuraghe, indipendente dalla specifica applicazione e configurabile dinamicamente per adattarsi alla specifica CNN, e un compilatore di CNN, che traduce la descrizione della CNN prodotta dal framework di machine learning in un programma eseguibile sulla MPU di Zynq utilizzando Neuraghe.


Recent improvements on image sensors, processing performance, power consumption, computer algorithms and machine learning, have elevated vision to a much higher level and over the next few years embedded vision will evolve into applications never seen before.
During this TechDay you will learn new technologies for embedded vision starting from image sensors to full embedded vision learning systems. The speakers provide a story that you will not find on the internet and new applications will be explained with 10 state-of-the-art vision solutions showcased in the afternoon session.
The TechDay will give answers on different topics, including:
– What is the future evolution on machine learning?
– How to create optimized vision solutions with little efforts in a short time?
– How to build power consumption efficient vision systems?
– What are the pitfalls when building a camera module?
– What is the trade off for implementing vision intelligence on the edge or in the cloud?
– How to optimize vision with poor environmental conditions?


08:30 Registration and coffee
09:00 Avnet Silica – Shaping the future, embedded vision megatrends
09:30 ON Semiconductor – Image sensor quality enhancement
10:10 TD next – Camera module design challenges
10:35 Avnet Silica – Alternative technologies and image sensor fusion
11:00 Break
11:20 STMicroelectronics – Outstanding graphical capabilities
11:45 NXP – Machine vision reshaped
12:10 Micron – Reliable memory & storage for secured embedded vision
12:35 Xilinx – Remarkable machine learning applications for industry
13:15 Lunch
14:15 Demos with individual discussions
17:00 End of demo sessions


Several solutions and tools will be demonstrated in the afternoon session.
You will be able to explore full embedded vision reference designs including camera drivers and application examples. The speakers and technical experts will be eager to answer your questions and to discuss your project.
Demonstrations include:
– Mixed reality with Microsoft HoloLens – every participant can try out the HoloLens
– Machine learning with deep neural network on a reconfigurable System-on-Chip (SoC)
– Real-time 3D sensing with Intel RealSense camera and computer vision SDK
– Configurable and modular imaging platform with audio and H.264 video compression
– Image fusion with different sensor technologies
– Modular reference platform for ADAS applications
– Usage of real-time computer vision libraries including OpenCV on SoCs
– Facial recognition and emotion analysis with cloud based cognitive services
– Visible Things: Vision enabled edge to enterprise industrial IoT platform
– Secured connected camera system
– 4K (Ultra HD) video reference design

Download Documentazione

FPGA accelerated Convolutional Neural Networks FPGA accelerated Convolutional Neural Networks
di M. Brian, A. Paccoia, C. Salati (1.537 KB)

NEURAghe: FPGA-based acceleration of convolutional neural networks NEURAghe: FPGA-based acceleration of convolutional neural networks
di Università degli Studi di Cagliari (443 KB)

Shaping the future for embedded vision with Deep Learning Shaping the future for embedded vision with Deep Learning
di Stefano Tabanelli (Avnet Silica) (2.707 KB)

TAGS: , , , , ,