Analytics over urban video streams is well suited for distributed computing across Edge, Fog and Cloud. Such streams are network intensive, making it is prohibitive to fully transfer them to the Cloud. Deep Neural Networks have achieved remarkable accuracy in image classification, but are computationally costly on just Edge devices. We propose ECHO as a big data platform to compose IoT dataflows and seamlessly distribute them across Edge and Cloud resources. In this demonstration, we illustrate the capabilities of ECHO for deploying several video analytics applications to support smart city use-cases.