3D Point Cloud Data (PCD) has been used in many research and commercial applications widely, such as autonomous driving, robotics, and VR/AR. But existing PCD generation systems based on RGB-D and LiDARs require robust lighting and an unobstructed field of view of the target scenes. So, they may not work properly under challenging environmental conditions. Recently, millimeter-wave (mmWave) based imaging systems have raised considerable interest due to their ability to work in dark environments. But the resolution and quality of the PCD from these mmWave imaging systems are very poor. To improve the quality of PCD, we design and implement MilliPCD, a "beyond traditional vision" PCD generation system for handheld mmWave devices, by integrating traditional signal processing with advanced deep learning based algorithms. We evaluate MilliPCD with real mmWave reflected signals collected from large, diverse indoor environments, and the results show improvements in the quality w.r.t. the existing algorithms, both quantitatively and qualitatively.