Introducing: TensorFlow™ Support for Neural Compute Stick
In July of this year, Movidius™ launched the world’s first USB-based deep learning inference tool – the Neural Compute Stick (NCS). This device was not just an industry first, but it was a special launch for Movidius, it being the first product officially launched as an Intel company.
We were extremely proud of the launch and subsequent reception of the device. During launch at CVPR 2017, we encountered three days of eager developers lined up to get hold of devices, and subsequent weeks of the system in the wild have already begun to yield some interesting uses of the device. For example, the medical startup Dr. Hazel recently utilized the Neural Compute Stick and caught major attention at Techcrunch disrupt by being able to flag skin blemishes with risk factors for cancer. While we cannot claim AI technology has cured cancer, it’s extremely heartening to see this technology being used in ways that’s beginning to solve such important problems.
In our developer community, we’ve seen intrepid developers porting applications to exciting new platforms and form factors, and numerous developers working together to get new networks supported – the community collaboration to get YOLO Tiny running on the device in short order is a great example of the budding teamwork we already see going on in our forums.
Despite the milestones and successes we’ve achieved over the last few months, there is much more we can do to make NCS the ultimate development tool for the AI revolution. That is why today we are pleased to announce that NC SDK now offers support for TensorFlow™. This was by far our most popular new feature request, and we hope this is a significant step in enabling more applications, experimentation and innovation.
Since its public launch in 2017, TensorFlow™ has quickly gained a large following in the machine learning space. The ability to develop and deploy within the same framework is a compelling reason to use the framework, and this is a philosophy we at Movidius also follow at the silicon level. With TensorFlow™ and Caffe support taken together, we hope that we’ve enabled the broadest swath of developer backgrounds possible with any two frameworks. We look forward to welcoming new TensorFlow™ users to our community, and enabling our existing users who have been waiting so patiently for this update.
Along with TensorFlow support, NC SDK v1.09.00 contains many new features such as:
- Tensorflow R1.3 support for tools (only on Ubuntu 16.04 LTS currently).
- New, unified, faster installer and uninstaller.
- Now supports complete SDK installation on Raspberry Pi.
- System installation of tools and API libraries.
- API support for Python 2.7.
- Source code included for API, for porting to other architectures or Linux distributions.
- Tools support for Raspberry Pi.
- More network support, see documentation for details!
- Virtual machine support.
*TensorFlow, the TensorFlow logo and any related marks are trademarks of Google Inc.