Dev Kit Weekly: Seeed Studio reComputer J1010 Edge AI Device
September 23, 2022
If you want your electronic system to “see” you’re going to need deep learning-based object detection capabilities. And if you’re running object detection algorithms in a remote safety application like wildfire smoke plume monitoring, you’re going to need a hardware platform with sufficient resources to capture and classify dangerous events as soon as they happen.
That describes Seeed Studio’s reComputer J1010 Edge AI development kit.
Seeed Studio’s reComputer J1010 Edge AI device is powered by a 4GB NVIDIA Jetson Nano module that houses not only a Quad-core ARM Cortex-A57 MPCore CPU, but also a Maxwell GPU with 128 NVIDIA CUDA® cores that brings 472 Gigaflops of performance to the table. It supports up to four cameras via 12 MIPI CSI-2 lanes and an 18 Gbps D-PHY 1.1 interface, and can encode 4K video channels at 30 frames per second while decoding them twice that fast.
But the real win here is that you get that performance at between 5 and 10 watts.
The Jetson Nano’s 4GB of 64-bit LPDDR4 memory shuttles data at 25.6 gigabytes per second, and is joined by an SD card slot and an additional 16 GB of eMMC 5.1 storage onboard the reComputer J1010. Other I/O includes an M.2 E-key for Bluetooth or wireless LAN cards, an RJ45 port that supports Gigabit Ethernet networking, an HDMI interface if you need to connect the kit to a display, and of course this USB Type-C port that delivers both data and power to the system.
Since the NVIDIA Jetson Nano in question is in fact a module, all of this connectivity is brought out through the Seeed Studio J101 carrier board, which measures in at 100 mm x 80 mm x 29 mm – a tick smaller than the kit’s overall 130 mm x 120 mm x 50 mm size. As you can see, that’s all contained within this aluminum enclosure, as is a passive heatsink which is designed to keep the Nano module cool in a range of environments without the reliability issues that come with a fan.
Courtesy the Jetson Nano, the reComputer J1010 comes with NVIDIA’s JetPack SDK pre-installed, providing out-of-the-box support for the entirety of the Jetson software stack like Jetson Linux, pre-trained neural networks, accelerated libraries, drivers, and oodles of other helpful tools, including numerous example projects focused on how to develop and deploy object detection in various use cases to help you get started.
Seeed Studio also provides a guide on how to train and deploy your own models using the YOLOv5 object detection algorithm in the open-source PyTorch ML framework. You can train the YOLOv5 model with your own custom image database, and since it only needs a few examples to get going, training time can be relatively fast. But if you want to skip ahead through that part, you can also train YOLOv5 algorithms with a public dataset like Roboflow’s wildfire smoke image database, available to license via Creative Commons.
Now you have everything you need (and more, honestly) to get your object detecting applications off the ground with all sorts of pre-trained models or your very own. If you’d like to get your own, the reComputer J1010 is available for purchase on Seeed Studio’s product page for $199.00 — or $179.00 each if you purchase 10 or more. Of course, as always, you can also enter the raffle below for a chance to win this kit for free. If you win, we’ll ship it right to your door, anywhere in the world, at no cost to you.
So, good luck in the raffle, good luck finding whatever it is you’re looking for (get it?), and we’ll see you on the next episode of Dev Kit Weekly.