3D digital shapes are now ubiquitous, from industrial digital twins and mock-ups to virtual environments and characters. Unfortunately, they are mostly represented as a set of basic geometric primitives, such as points and polygons, without much global coherence. How can we recognise an object or its parts from such low-level information? This is the purpose of 3D shape understanding, which I will introduce in this talk with a specific focus on plants and trees. As an example, I will present a new citizen science app which allows one to compute various urban tree features from a single smartphone video.