This work considers a scenario in which an edge server collects data from Internet of Things (IoT) devices equipped with wake-up receivers. Although this procedure enables on-demand data collection, there is still energy waste if the content of the transmitted data following the wake-up is irrelevant. To mitigate this, we advocate the use of Tiny Machine Learning (ML) to enable a semantic response from the IoT devices, so they can send only semantically relevant data. Nevertheless, receiving the ML model and the ML processing at the IoT devices consumes additional energy. We consider the specific instance of image retrieval and investigate the gain brought by the proposed scheme in terms of energy efficiency, considering both the energy cost of introducing the ML model as well as that of wireless communication. The numerical evaluation shows that, compared to a baseline scheme, the proposed scheme can realize both high retrieval accuracy and high energy efficiency, which reaches up to 70% energy reduction when the number of stored images is equal to or larger than 8.