IoT Smart Camera Landscape
Qualcomm Vision Intelligence Platform + Microsoft Azure
Qualcomm just announced a device aimed at developers in partnership with Microsoft and their Azure platform. Microsoft and Qualcomm are coming out with a Vision AI Developer Kit which you can signup and learn more about now. From what I can see they’re offering a cloud based Azure Machine Learning Service coupled with their Azure IoT Edge Platform to take your cloud trained AI model and run them on cross platform edge devices, like smart cameras. All of this along with what appears to be a dockerized deployment of your cloud trained models to be run on device.
All of it powered by Qualcomm’s new Visual Intelligence Platform and Snapdragon Neural Processing Engine (NPE-yet another acronym). They have been talking about these types of devices since since 2015. The details I can find on the internet are a little fuzzy on to whether the NestCam IQ is powered by one of these newer chips or the older ones repurposed to do AI on device. Either way it is great to see other chipset manufacturers and software service providers involved in innovating smart device development.
Amazon’s AWS Deep Lens
This is yet another smart camera developer kit to hit the market and it is great news for GreenThumb IO. I’m excited to get my hands on the Amazon Deep Lens which will be release June 14, 2018. Amazon has teamed up with Intel and their Atomprocessor, which they quote here can process 100 billion floating-point operations per second (GFLOPS). While it’s not out yet they’re already blogging about it here. It looks like they’re giving you the power of AWS Greengrass and Lambda on device. This is going to enable cloud power AI solutions to break free of the chains of centralization and go to edge.
Google’s AIY Vision Kit
I’ve mentioned on my first blog post that the Alphabet is ahead of the curve again with the Google AIY Vision Kit. Launched November 30, 2017, the AIY Vision Kit piggybacks off an ARM CPU powered Raspberry Pi, with a vision bonnet that features a Movidius Chipset. This VPU coprocessor is used to execute TensorFlow models on the device while the ARM CPU is used for program logic. It’s encouraging to see so many big companies enabling developers with these proprietary devices with fully programmable with open source software.
It would appear the giants are gearing up for next revolution of smart devices. It seems to be a competition for operating platforms and services. As someone with aspirations of my own it makes me wonder which one of these companies is ready to start partnering with startups to encourage growth in this area.
NVIDIA’s Jetson Platform
the embedded platform for autonomous everything. — NVIDIA on Jetson
NVIDIA, which I mention in my last blog, also has a developer kit of their own with the Jetson Platform, the embedded platform for autonomous everything. Its powered by the awesome Tegra 4.0 chipset which boast a beefy ARM CPU as well as 256 CUDA enabled GPU cores. NVIDIA has been powering both the crypto and artificial intelligence hypes and it’s been reportedly taking a toll on their stocks. Regardless NVIDIA is the clear leader in GPU/TPU technology as well as open source software. CUDA has paved the way for modern day deep learning frameworks such as TensorFlow, Caffe, Theano, PyTorch, and so on.
Where do you end up once you have a marketable smart camera product? — me just now!
So what does it mean for the ambitious developers and entrepreneurs out there. Personally i think it’s inspiring, but if you build something cool with any one of these platforms you have to ask yourself a question. Where do you end up once you have a marketable smart camera product? I’ve submitted an application for GreenThumb IO to join the NVIDIA inception program, you can read my application below. Enjoy!
GreenThumb IO is developing a Smart Camera platform for commercial cannabis cultivators using Blockchain-powered decentralized Artificial Intelligence.
Utilizing over 1000 cameras per acre, powered by the NVIDIA Tegra platform, we will actively monitor and learn at the commercial scale of hundreds of thousands acres. We will use Deep Learning techniques to not only enhance the Computer Vision that powers the analytics, but also cultivate an artificially intelligent agronomic understanding of key visual growth indicators.
We’re starting out with TensorFlow trained models that can infer and mask leaf area, bud area, stress, and harvestability at a millimeter per pixel resolution. These enormous high resolution datasets, coupled with the power of NVIDIA’s distributed CUDA framework and GreenThumb IO’s distributed AI, will create an unprecedented super human understanding of cannabis agronomy.
The system is designed to transmit metrics from the Computer Vision Models to blocks. The cameras GPUs participate in solving blocks when they’re not capturing and analyzing their FOV. We hope to run the TensorFlow inference on the ARM CPU. This is a practical and scalable blockchain solution that will limit if not virtually eliminate cloud servers.
With NVIDIA, GreenThumb IO will be the most advanced Decentralized AI Smart Camera Platform.
Justin Bowen, GreenThumb IO