Shot boundary detection (SBD) is an important first step in many videoprocessing applications. This paper presents a simple modular convolutionalneural network architecture that achieves state-of-the-art results on the RAIdataset with well above real-time inference speed even on a single mediocreGPU. The network employs dilated convolutions and operates just on smallresized frames. The training process employed randomly generated transitionsusing selected shots from the TRECVID IACC.3 dataset. The code and a selectedtrained network will be available at https://github.com/soCzech/TransNet.