Triangulating Precision: A Comparative Study of Manual and Automated Annotations with YOLO,Azure Custom Vision and Grounded SAM on a Customized Data set for creation of a product for safety of recycling industries
Master thesis
Permanent lenke
https://hdl.handle.net/11250/3123481Utgivelsesdato
2024Metadata
Vis full innførselSamlinger
Sammendrag
This thesis centers on the imperative task of detecting and segmenting batteries within therecycling industry, addressing the need for an efficient and accurate solution. The primarygoal is to conduct a comprehensive comparison of state-of-the-art methods to discern themost suitable approach for the specified task. The comparative analysis extends to bothmanual and auto annotations, where manual annotations employ Roboflow, feeding datainto Ultralytics’ YOLOv5 and YOLOv8 models, as well as Azure Custom AI. Auto annotationsleverage Grounded SAM and GroundingDINO, with Grounded SAM data integratedinto YOLOv8 and YOLOv5. Remarkably, YOLOv8, combined with manual annotationsfor the custom battery dataset, demonstrates significant success. The thesis concludes byselecting the most effective method and enhancing a health dashboard based on the chosenmodel, providing a comprehensive and practical solution for the recycling industry’s batterydetection and segmentation challenges.