Production conditions change – lighting, process parameters, and new product variants can all challenge existing machine vision systems. In this demo, Agnes Weinhuber from MVTec shows how Continual Learning helps deep learning classification models adapt directly on the edge. The system detects a new object class using out-of-distribution detection, saves sample images, and retrains the model in seconds using only 5 to 15 images – without needing a GPU server or external development PC. Watch the video [Picture from video by MVTec Software on youtube.com]
