Douglas Bagnall: Cloud Shape Classifier

  • ©, Douglas Bagnall, Cloud Shape Classifier

Artist Statement:


    To save people time in the search for interesting clouds, a computer watches the sky all day, every day. Viewers can interact with the computer via the Internet. It establishes individual relationships with each person, developing an idea of the kinds of clouds they like. When they return to the site they are shown the clouds that have passed that would have been their favourites. Their reactions help refine the computer’s idea of their taste. It will also show the greatest clouds by popular opinion.

    The computer and camera will be in Wellington, New Zealand.

    The cloud shape classifier consists of a computer and a camera, which is tilted skyward. Every few seconds, an image of the sky is captured. Each image is analysed and classified according to shape, texture, and colour, and saved to disk.

    The images are presented via a website. Visitors to the site will see a selection of the most popular clouds. If they choose to identify themselves, they will be shown selections of clouds from which they can choose favourites. As they make selections, the machine will attempt to show them new clouds that ought to be to their taste.

    Internally, the machine views its images as points in a vector space with several hundred dimensions. Similar clouds will be close together in this space, and the coordinates within it will be fed to neural networks which will learn the preferences of each person.

    The cloud selection interface is likely to be an array of nine images. Clicking on any of the outer 8 images re-centres the array on that image, as a way of navigating through the vector space. Clicking on the central image enlarges it to fill the screen, and casts a vote, modifying the neural network in its favour.

    The most popular clouds will be determined by committees of these neural networks. Therefore a cloud which has not previously been shown to a human may be selected as the most popular.

    The machine is going to be installed at the Enjoy Public Art Gallery in Wellington, New Zealand, at around the time of ISEA2006. There will be a projection in the Enjoy Gallery, as well as the web interface. The Enjoy people would be happy to share the exhibit with ISEA/ ZeroOne San Jose — given that it is primarily web based, this is easy. Perhaps a camera could be installed in both cities. Alternately, it could be installed in San Jose just before or after the Wellington show.


Technical Information:


    The software will be written in C and python, and run on Debian Linux.


Category: