On September 13th, 2017, VisionHack, the first international hackathon on artificial intelligence and computer vision, concluded. This momentous event was organized by NUST MISIS and Cognitive Technologies under the support of State Corporation “Bank for Development and Foreign Economic Affairs (Vnesheconombank)” and the Skolkovo Foundation. The top three winners included two teams from Lomonosov Moscow State University, while the third hailed from MIPT. Two teams from NUST MISIS, one team from Skoltech, one from Kazan Federal University, and one from Innopolis were included in the competition’s top 10. The final slot in the top 10 went to the team from UPC (Spain).
For three days, 27 teams from the 22 leading Russian and foreign universities such as University of Cambridge (UK), Massachusetts Institute of Technology (USA), Arizona State University(USA), Polytechnic University of Catalonia (Spain), Peking University (China) and seventeen more went about solving the task of creating their own intellectual subsystem for driver assistance and an ADAS (advanced driver assistance system) capable of automatically detecting various road events.
DoubleA Team from Lomonosov Moscow State University took 1st place, winning the amount of $12,000. Deep MIPT from Moscow Institute of Physics and Technology took 2nd place. GML Vision, another team from Lomonosov Moscow State University, took 3rd place.
The task for the hackathon`s on-site competition consisted of 5 subtasks of different levels. Participants had to determine the presence (or absence) of bridges & tunnels, zebra-stripe crosswalks, speed bumps, road entry/exit signs forvarious communities, and road cleaning & work.
Despite theirclear victory in the point total, DoubleA Team from Lomonosov Moscow State University didn`t manage to win the jackpot. Andrey Belyaev, captain of DoubleA Team, said that they went withthe method of neural networks training, “Many participants thought that they could apply many good approaches to the solution of the assigned tasks without neural networks. However, we believe that the future is neural networks, and did our best to prove it through a personal example. That`s why during the hackathon we developed the necessary architecture to configure the work of neural networks for the tasks, and as a result have shown the best results in almost every type of task [here]. Only in the case of the speed bumpdid a few teams offer a solution better than ours”.
4th place was awarded to the Tapochki team, composed of individual participants from the preliminary selection stage—students from Lomonosov Moscow State University and Saint Petersburg State University, as well as Nikita Krylov, a school student from Moscow school № 1207, who was the youngest participant of the hackathon.
“Two patterns can be seen in the solutions,- noted Darya Krokhina, spokesperson for the hackathon and coordinator of the testing system,— some of the teams worked on neural networks training, while others focused more on computer vision in its purest form. They just looked for objects which have the same pattern in the frame. It is a more low level solution but it works faster, so in conditions of limited resources (48 hours) such decisions had the chance to succeed. The ‘Bridge’ was the simplest task for participants. On the evolutionary selection several teams even passed it with a full score. However, the retraining took place, and the final results were lower. The task for the determination of an entry or exit sign to a given community was the most difficult one for the teams. The best solution determined only half of these events, and some teams even received penalty points as the testing program fined them for the false definition of an event.”
A hackathon of such a scaleas this, predominantly dedicated to computer vision and unmanned vehicles, was held for the first time not only in Russia, but worldwide as well. The hackathon`s total prize fund exceeded $30,000, which is a record for such events. According to the terms of the hackathon, all intellectual property created during the event belongs to the teams.
“The victory of Russian programmers on the subject of artificial intelligence for autonomous cars can be called historic in every sense of the word. It once again emphasized the high level of our national schoolingfor artificial intelligence and its credibility around the world”, said Olga Uskova, president of the Cognitive Technologies group.
According to Alevtina Chernikova, Rector of NUST MISIS, “Events like the VisionHack hackathon allow NUST MISIS students to form and demonstrate qualities which are necessary in the modern world—the ability to work in a team, to take responsibility, and to solve non-routine tasks in a short amount of time. It is important to note that the hackathon`s participants were given real tasks to solve, and they managed to cope with them really well.”
The results of the participants’ work can be viewed on the hackathon`s website.