Summer Interns at Groke Technologies

Summer is usually the time when companies get summer trainees in and we at Groke Technologies are no exception to this. This summer we hired again trainees to our R&D team, most of them were working on different machine learning solutions and trialling some new approaches which we could use in our system and development environment in future.

Semantic Segmentation

For example Oksana Havryliuk worked on semantic segmentation. It is a deep learning algorithm that labels each pixel ?and segments in a picture into different areas like sea, vessel, sky and so on, but let’s have Oksana tell in her own words what she did:

“I researched and experimented on the semantic segmentation model to provide better distance calculation for ships. My duty was to do research on available pretrained segmentation model from Pytorch and perform experiments and evaluate each of them on Groke dataset.”

Purpose why we gave her this assignment was that we wanted to experiment different methods for calculating the distance to detected objects. Below you will see example of the segmentation.

The main purpose of the segmentation is to differentiate different objects in the picture, as seen above the different objects (sea, vessels, sky, land) are segmented with different colours. This makes interpretation of objects easier for the computer models.

Oksana made also a video of her typical day at Groke, you can check it out here: https://youtu.be/kvDb2xKTiwA

Working on Learning Database

Samir Huseynzade worked on our machine learning database and on our annotation tool. One specific task included trialling machine learning framework called GAN Network, and CycleGAN to be precise. Generative adversarial network (GAN), is a class of machine learning systems. Two neural networks contest with each other in a game (in the sense of game theory, often but not always in the form of a zero-sum game). Given a training set, this technique learns to generate new data with the same statistics as the training set. For example, a GAN trained on photographs can generate new photographs that look at least superficially authentic to human observers, having many realistic characteristics. Then what CycleGAN does it then also does the image creation back to original format. Much used example for this is the zebra and horse example where horse images are created from zebra images and the vice versa.?

UI Testing

Another topic worked on during summer was to try out new approach for our user interface testing. This assignment was carried out by Alberto Carlos Hernández , his work included trying the AltUnity tester and building relevant test environment to run trial tests. The purpose of this work was to look into novel testing technologies which would automate the testing process. In principle with AltUnity we can test our user interface and its functionalities automatically. For example Carlos built test where transitions between different menus in the user interface was automatically tested without the need of tester to push the different menu symbols & buttons.

Groke Active with Local Schools

Groke Technologies have been a voice, not only in universities but also in local highschools promoting engineering and other STEM field as a career path. During one of those visits student Inka Rekonen took an interest for our company and we had a pleasure to have her as a summer worker on the annotation team. During her traineeship she annotated over 4000 maritime images collected by Groke, improved our guidelines for annotation and by feedback affected our future development of annotation process.

During summer we had four trainees working with us, but we don’t only employ students during summer but in addition to traineeships we also provide students possibility to earn additional money during their studies by doing annotation work for us as freelancers. Annotation work means that they label pictures with detailed information what is seen in the picture utilising a special tool we have made for this purpose.?

There is plenty of annotation work to be carried out as our image database includes over 11 Million images so we are always looking for new annotators to work with us. If you are intersted in joining our annotation team (only available in Finland at the moment) as freelancer contact: [email protected]

The results from all work carried out were really promising and it has to be said that again this summer we were really lucky to get such capable students to work with us!

Article originally published in our exclusive customer newsletter in October 2022