Активне навчання у задачі виявлення об’єктів
Loading...
Date
2022
Authors
Ронська, Дарина
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Active Learning allows to spend less time on data labeling which is vastly beneficial in Computer Vision with continuously growing number of datasets and its images. It is achieved by smarter strategy than random one to queue the images for labeling that allows to give most informative images to the model first. In this work state-of-the-art Multiple Instance Active Learning for Object Detection (MI-AOD) method is improved by the changes in its uncertainty function which corresponds for informativeness of the image. Also, the statement of MI-AOD authors about its usage for noisy images filtering is proved.
Description
Keywords
MI-AOD: Multiple Instance Active Learning for Object Detection, рroposed metrics for uncertainty re-weighting, metrics comparison and results, motivation to use MI-AOD for blurred pictures filtering, магістерська робота