Hlavní stránka Info portálu ZČU

The eighth edition of the #AimtecHackathon programming competition took place at the beginning of March this year, with the subtitle When Code Helps. The Faculty of Applied Sciences of the University of West Bohemia in Pilsen (FAS) was represented in six teams. The winner was the Vision team, which also won the Audience Award.

Last year, members of the Vision team, with almost the same composition, took third place at the hackathon. "This year, we could already draw on last year's experience. We knew what to primarily focus on, how to divide and organize our work, what to avoid, and what to work on," says Marek Hanzl. In this year's challenge, the students chose a project that would help the blind "see." This is the ImageChat app, which works well indoors and outdoors. "Blind people move more confidently in their home environment than outdoors. Our app is a kind of supplement or assistant for them," adds Jakub Herman.

How does the app work? The blind person downloads it to their mobile phone. If, for example, he wants to know if there is an obstacle in front of him during a walk, he will take a picture of the area and then ask. ImageChat can not only recognize the objects that are in the photo, but it can also recognize that a particular object is not there. They plan to continue working on the app and improving it in the future. They are currently thinking about which path to take. "We are considering whether to go the commercial route or the charity route. The developed model has to run on some external server, and of course that costs money. We're figuring out how to finance the whole thing so that it makes sense," explains Daniel Zierl. "If we go the commercial route, we must fine-tune the functionality. We would introduce ads and tokens. If we go the non-commercial route, we have to figure out how to finance the server part," adds Daniel Zierl.

ImageChat currently has two modes, and it's a question of how to tweak the polling. In the first mode, the questions can be predefined, as they are very likely to be repeated. In this case, it would be enough to select a question from the menu, send it to the server, and get the answer immediately. In the second mode, the blind person asks questions using his own questions. In this variant, the model should remember the history and context of the conversation and should be able to follow up meaningfully in the conversation.

"Our model benefits a lot from context. We can tell him you're in a city in the Czech Republic, you're outside, and he can work with that input. So when we take a picture of a street, the model knows that there are cars, curbs, etc. But it doesn't need to be able to distinguish details. For example, I can take a photo from a café where they have a palm tree inside, and the model can evaluate that I'm on the beach," explains Marek Hanzl.

Interestingly, the youngest team member, Tomas Zierl, the brother of one of the FAV students, is only in primary school. His task was to develop the graphic design of the website. And as the other Vision team members confirm, their youngest colleague did a great job. 

ImageChat can be downloaded from the Google Play Store. The app is in the early stages of development, and testing is ongoing.