THE MARKER: ARTIFICIAL INTELLIGENCE-ASSISTED IMAGE MARKING TOOL
Keywords:
Machine Learning, Computer Vision, LabelingAbstract
This undergraduate thesis aims to develop The Marker, an AI-assisted image-annotation tool designed to create customized datasets for computer-vision applications, grounding the study in concepts of artificial intelligence, machine learning, deep neural networks, and ergonomics while emphasizing the importance of image annotation in building effective computational models and the physical impacts associated with repetitive tasks such as RSI, WMSDs, and Computer Vision Syndrome. The applied methodology involved developing a modular application composed of a React graphical interface, processing modules in Rust, execution of the Segment Anything Model (SAM) via Python scripts, and secure storage with AES-GCM encryption; experimental tests were conducted to evaluate accuracy, interference time, the number of manual interactions required, and system performance across different image resolutions. The results indicate that the tool significantly reduces manual effort by automatically suggesting segmentation points, operates offline and on lower-powered machines, provides an improved ergonomic experience, and shows strong potential to accelerate the collaborative creation of visual datasets.
