Faces in Event Streams dataset contains 689 minutes of recorded event streams, and 1.6 million annotated faces with bounding box and five point facial landmarks. Tha paper presents the dataset and corresponding models for detecting face and facial landmakrs directly from event stream data.