Bookmark and Share

CEVA Demonstration of Real-Time Deep Learning-based Object Recognition

Marc Evans, from the Customer Solutions Group at CEVA, demonstrates the company's latest embedded vision technologies and products at the May 2016 Embedded Vision Summit. Specifically, Evans demonstrates the AlexNet open source network running on CEVA's XM4 vision processor. The network includes 24 layers and is trained using the Caffe deep learning framework. The demo includes a 1080p webcam connected via USB to an i.MX6 host platform, connected via PCIe to the CEVA-XM4 FPGA Development Board. A single CEVA-XM4 vision processor core is implemented on the FPGA, running at 40 MHz. In an ASIC running at 800 MHZ and above, the object recognition time will be 20-30 times faster than what's seen in this demo.