Latest in the series of critical design projects by Shanghai design and research studio Automato, TraiNNing Cards is a set of 5000 training images, physically printed and handpicked by humans to train any of your machines to recognise first and favorite item in a house: a dog.
Recently we hear a lot about Machine Learning, Deep Learning, and Neural Networks as being a revolutionary way for computers and machines in general to sense and recognise the world around them. Security cameras, home robots and other “intelligent” machines will have to recognise not only us but objects, pets and other inhabitants of our life with a training process that is hidden and done digitally through a generic set of images gathered from the web. But how much does it take for a “machine” to learn something? Well, around 1.2 meters and 8 kg of images.
For a neural network (NN) to be able to recognize something as basic as a dog (or any kind of objects) it needs to be trained with a certain number of images representing that object as well as images that do not represent that object. Based on NN specialists best practices, 5000 images is the right number for a neural network to start having some idea about the object you have trained it with. Automato’s TraiNNing Cards is a set of 5000 training images that includes both “true” images (dogs) and “false” images (non-dogs) with easy to read QR codes for your machine to clearly identify what is a dog and what is not a dog.
The Basic set 001 #dog is sold for “only 5000$” (presumingly with copyright licence for you machine’s piece of mind), but should you need training cards for another object, Automato can create custom sets on request. Additionally, they will personally curate the imagery for you and produce a set with their local machine-lovers producers in China.