Marek Vajgl

We present a conceptually simple yet powerful and general scheme for refining the predictions of bounding boxes produced by an arbitrary object detector. Our approach was trained separately on single objects extracted from ground truth labels. For inference, it can be coupled with an arbitrary object detector to improve its precision. The method, called BBRefinement, uses a mixture of data consisting of the image crop of an object and the object’s class and center. Because BBRefinement works in a restricted domain, it does not have to be concerned with multiscale detection, recognition of the object’s class, computing confidence, or multiple detections. Thus, the training is much more effective. It results in the ability to improve the performance of SOTA architectures by up to two mAP points on the COCO dataset in the benchmark. The refinement process is fast; it adds 50–80 ms overhead to a standard detector using RTX2080; therefore, it can run in real time on standard hardware. Finally, we show that BBRefinement can also be applied to COCO’s ground truth labels to create new, more precise labels. The link to the source code is provided in the contribution.

We present a scheme of how YOLO can be improved in order to predict the absolute distance of objects using only information from a monocular camera. It is fully integrated into the original architecture by extending the prediction vectors, sharing the backbone’s weights with the bounding box regressor, and updating the original loss function by a part responsible for distance estimation. We designed two ways of handling the distance, class-agnostic and class-aware, proving class-agnostic creates smaller prediction vectors than class-aware and achieves better results. We demonstrate that the subtasks of object detection and distance measurement are in synergy, resulting in the increase of the precision of the original bounding box functionality. We show that using the KITTI dataset, the proposed scheme yields a mean relative error of 11% considering all eight classes and the distance range within [0, 150] m, which makes the solution highly competitive with existing approaches. Finally, we show that the inference speed is identical to the unmodified YOLO, 45 frames per second.

The continuous monitoring of a shore plays an essential role in designing strategies for shore protection against erosion. To avoid the effect of clouds and sunlight, satellite-based imagery with synthetic aperture radar is used to provide the required data. In contrast to standard model-driven methods, we present a deep-learning-based approach to detect coastlines in such data. We split the process into data preprocessing, model training, inference, ensembling, and postprocessing and describe the best techniques for each of the parts. To deal with a small training dataset, we propose a novel multi-sample mosaicing augmentation that helps the deep neural network models to reduce overfitting during training. Our solution has been validated against the real Global Positioning System location of coastlines during a worldwide competition organized by Signate and Japan Aerospace Exploration Agency, where it was runner-up among 109 teams from the whole world.

During our research, we have presented a novel approach for multi-focus image fusion based on new image sharpness measures built on F-Transform. We have already presented that F-Transform based sharpness measure subjectively give good results in edge detection and sharpness evaluation. In order to continue with our research, we need to prove that such F-Transform based measures are generally usable as sharpness measures in image processing. So we aimed at our measures trying to evaluate their stability and robustness against noise. In the contribution, we present theoretical background and properties requested to be satisfied by sound sharpness measures. Then, we presented one previously introduced and one novel F-Transform based sharpness measure, and according to the presented properties, we verified that the proposed measures are fulfilling stated properties. The verification is done over experimental and real-life images and is evaluated referring to the standard simple sharpness measures. The result is that the both F-Transform based proposed measures supports presented properties and can be generally used as image sharpness measures.

Measures of image sharpness are essential in the field of image processing. Their application is a key for many more complex algorithms. There is a big amount of different image sharpness measures. We have repeated their expected properties according to previous research and checked if a selected set of image sharpness measures fulfill the required conditions in artificial and real-life experiment.

The contribution introduces a research in the field of Moodle system plugin with the ability to process an evaluation of Java programming homeworks in an automated way. An automated assessment consists of two parts – plagiarism evaluation comparing homeworks against each other, and an automated correctness evaluation of the homework. The plugin has to implement a security measures in a form of sandbox feature to ensure Moodle system security and stability.