Point cloud based place recognition is still an open issue due to thedifficulty in extracting local features from the raw 3D point cloud andgenerating the global descriptor, and it's even harder in the large-scaledynamic environments. In this paper, we develop a novel deep neural network,named LPD-Net (Large-scale Place Description Network), which can extractdiscriminative and generalizable global descriptors from the raw 3D pointcloud. Two modules, the adaptive local feature extraction module and thegraph-based neighborhood aggregation module, are proposed, which contribute toextract the local structures and reveal the spatial distribution of localfeatures in the large-scale point cloud, with an end-to-end manner. Weimplement the proposed global descriptor in solving point cloud based retrievaltasks to achieve the large-scale place recognition. Comparison results showthat our LPD-Net is much better than PointNetVLAD and reaches thestate-of-the-art. We also compare our LPD-Net with the vision-based solutionsto show the robustness of our approach to different weather and lightconditions.