"It was a pleasant surprise. I wasn't expecting it, which makes me even happier that my thesis received this award," says Juliána Mikešová.
The thesis focuses on the design of a system for the efficient expansion of datasets used for image object detection. It addresses the challenges of automatic object detection, compares models based on convolutional neural networks, and discusses the requirements for datasets used to train these models.

As part of the thesis, Juliána designed a system for expanding a vehicle detection dataset using the Grounding DINO transformer for semi-automatic data annotation. The generated annotations are then presented to the user for review and confirmation, significantly simplifying and speeding up the dataset creation process. The thesis also includes comprehensive user documentation.

"The outcome of this thesis is a unique system for the efficient expansion of image object detection datasets, which is currently being used in several projects at our department. Its efficiency lies in leveraging the output of the existing detection model, allowing the dataset to be expanded only with images on which the current model fails," says the thesis supervisor, Jaromír Továrek, Ph.D.
Juliána Mikešová will continue building on the results of her master's thesis in her doctoral studies.

"During my Ph.D. studies, I plan to focus on image processing using data from emergency services and integrating large language models into the process," she says.