Computer vision specialist EyeCloud has launched what it describes as "the first open AI vision appliance:" the OpenNCC, a fully-integrated system for rapid deployment — and built on top of open source software.
"We want to change the way people look at AI development," claims EyeCloud's Po Yuan. "Before, you had to use powerful machines, multiple components, and a lot of time to develop and train AI algorithms. Now, all of these can be achieved with our AI vision appliance that has it all."
"OpenNCC is a production vision system with the AI vision hardware and software for real-time ML application processing; tightly-integrated with high-quality vision camera, ISP, video processing and communications infrastructure."
The OpenNCC hardware is built around Intel's Myriad X vision processing unit (VPU) accelerator, offering up to four teraflops of compute performance, with 1GB (8Gb) of LPDDR4 memory and 16MB of SPI flash storage. The compute hardware is positioned behind a 2.3 megapixel camera sensor with a global shutter running at up to 120 frames per second and compatible with C-mount camera lenses; the OpenNCC IR+, meanwhile, also includes a thermal image sensor.
Which model you get will depend on your use case: The OpenNCC USB includes a USB Type-C 3.1 port for connection to a controlling PC, while the OpenNCC IPC is an independently-operable network device with Power over Ethernet (PoE) support. The OpenNCC IR+, meanwhile, is also designed for network use, and is the only model in the family to include a video output courtesy of an HDMI port.
The OpenNCC family is designed for use with OpenVINO, supporting the ONNX, TensorFlow, Caffee, MXNet, and Kaldi frameworks and programmable in C, C++, and Python. Pricing starts at $299 for the OpenNCC USB, $399 for the OpenNCC IPC, and $1,495 for the OpenNCC IR+ — but pricing will rise for all models when the initial Early Bird rewards expire.
The OpenNCC crowdfunding campaign is now live on Kickstarter, with hardware delivery scheduled before the end of the year.