 The paper presents a new 3D point cloud classification benchmark data set with over 4 billion manually labeled points. The data set is meant to be used as input for deep learning methods, specifically convolutional neural networks, CNNs. The authors discuss first submissions to the benchmark that use CNNs and show remarkable performance improvements over state-of-the-art. The paper aims to close the data gap for 3D point cloud labeling tasks by providing a massive data set consisting of dense point clouds acquired with static terrestrial laser scanners. The semantic 3D.NET data set contains eight semantic classes and covers a wide range of urban outdoor scenes. The authors describe their labeling interface and show that the data set provides more dense and complete point clouds with much higher overall number of labeled points compared to those already available to the research community. They also provide baseline method descriptions and comparison between methods submitted to their online system. The paper hopes that semantic 3D.NET will pave the way for deep learning methods in 3D point cloud labeling to learn richer, more general 3D representations. This article was authored by T. Hackel, N. Savanov, L. Ladiky and others. We are article.tv, links in the description below.