Autolabeller tool based on the original LabelImg by adding the Autolabel function to it using the trained object detection models. This function helps reducing the human resources as well as labelling time by 3 times by empowering the object detection model.

Successfully released the software to monitor the medication usage behavior of the patients. This software is a combination of deep learning object detection models trained on real video data collected from hospitals and volunteers to detect objects in videos and LSTM models to analyze the action by considering videos as time-series data. The object detection models are built upon Tensorflow Object Detection API while action recognition models are built from scratch and updated daily using Kubeflow on Azure. The accuracy was more than 95% and the model was brought to cooperate with several hospitals in Korea.