Turn-key PCB assembly services in prototype quantities or low-volume to mid-volume production runs

Efficient GPU based embedded platform for intense computation

There are plenty of credit card size computer modules that usually are based on ARM or x86. They are great for general purpose tasks like running operating systems, playing games and running apps. But when you need to squeeze a bit more out of them they appear not very powerful. One group of intense tasks would be running deep learning algorithms by using neural nets or performing other machine learning tasks. Along with great performance we also are looking for low power solutions especially for portable applications like self aware robots. Nvidia recently launched great platform called Jetson TX1 that just fits this category. It is based on 1 TFLOP/s 256-core Maxwell GPU, 64-bit ARM A57 CPU featuring 4GB of DDR4 RAM and 16GB of eMMC flash. Along with great performance it only takes 10W. As we already may know GPU is great in image processing and classification so it outperforms general purpose CPUs in these tasks. When compared to Intel core i7-6700K and taking energy consumption in to account Jetson TX1 showed over 6 times better performance. Of course… Continue reading

Face recognition with Raspberry Pi

Face recognition is interesting and really important topic. Digital cameras already does this to help better focus on a subject. Face recognition also can be used in areas such as robotics, computer-human interaction, security systems and so on. Boris Landoni has been working on face recognition with Raspberry Pi + camera module. He wanted to adapt simple, robust algorithm that would detect human face by determining parts like face, eyes, nose lips. The recognition method is based on so called “Haar feature cascade” which allows rapidly to identifying objects by means of cascade consecutive combinations of simple features. In other words features from images are extracted by using Haar wavelets. Face recognition program runs on Python that uses SimpleCV library for algorithm training. So far this is not an end product, but rather a demonstration of complexity of problem. Continue reading

RC car got brains

Many of you probably took a chance to complete free machine learning course led by Andrew Ng. If not don’t hesitate to register for 2012 ML course. David was taking this course and already came up with his own project. He took RC toy car and transformed it in to self driving system. Whole setup consist of RC car, Android phone that takes pictures and sends to a server running on Laptop. Here images are passed though neural network. Once car control signal is calculated it is passed through Arduino to RC control keypad which adjust cars next movement (right, left, forward or backwards). Continue reading