System developers CodeMill AB are joining MICO as a technology-transfer partner, and will as such help disseminate project results to a wider audience. The Swedish company services media-oriented customers on an international market, typically building on open-source platforms with a focus on usability. Recent projects include a suite of metadata-extractors for the Vidispine platform, and a video publishing system tailored for Guardian News and Media. The company was founded in Umeå 2007, and still has its head office there. Today, the company consists of just over 30 employees with levels from junior to senior computer scientists, interaction designers and even PhDs.

“Our customers are continually challenging us to find new and better ways to manage their content”, says CEO Rickard Lönneborg, “and we believe that the MICO plattform shows great potential as a unifying framework.”

By participating in MICO, CodeMill wants to contribute to a much appreciated research initiative, and to encourage knowledge exchange in the field of video analysis. This is in line with the company’s general policy of acting in the intersection of Industry and Academia. CodeMill has previously participated in the EU-project SATIN on user-driven development of mobile applications, and has on-going collaborations with researchers and start-ups in the field of media analysis.

“The availability of more and better data sets, together with advances in machine learning, is driving progress. We expect that many relevant forms of data analysis will cross over from the purely theoretical sphere into the real world.” adds CTO Johanna Björklund. “Emotion recognition is a prime example, we see lots of interesting things happening there.”

CodeMill believes that MICO can accelerate the progress even further, as it allows data analysis to be made with respect to many different aspects of a media asset. It may e.g. be difficult to determine who appears in a low-resolution video based on the visual component alone, but by adding in speaker recognition and extracting proper names from associated comment fields, precision can be improved.

“The beauty of it is that the abstractions used in many machine-learning models allow for different kinds of features to be blended freely. Such is for instance the case for support vector machines and neural networks. Now we would like to see this principle applied in practise.”, Björklund concludes.