In this work, an overview of the submitted run to TRECVID 2019 by ITI-CERTH is presented and more specifically, for the task of Activities in Extended Video (ActEV). Towards this objective, we deployed a state-of-the-art architecture for the human action recognition problem and with the application of an encoder-decoder model, we extract a threshold for every activity in order for the framework not only to recognize activities but also to identify them in extended videos.