Object Detection for Virtual Assistant in Cryolaboratory Based on Detectron2 Framework

In this work, the previously proposed structural scheme of a virtual assistant is intended for training biomedical engineers and other specialists in working with specialized high-tech equipment, using a cryolaboratory as an example. The necessity of incorporating a specialized detection stage, which should consider the specific visual characteristics of the particular laboratory, has been demonstrated. Three models were trained on the R-CNN architecture (based on the Detectron2 framework) to create a virtual guide for a 3D cryolaboratory. The metrics of the obtained models were analyzed, and their suitability for specific tasks was evaluated. The obtained results indicate sufficient detection performance of the cryo-laboratory components, allowing the use of the developed models to create a specialized virtual assistant.