A three-dimensional (3D) interactivity system automatically and dynamically generates shape-conforming and computationally efficient colliders for detecting collisions with automatically differentiated features represented by different sets of points in a point cloud. The system selects a set of points that represent a particular feature of a 3D object, decimates the set of points to a subset of points that represent an approximate shape of the particular feature with fewer points than the set of points, and generates a collider with the approximate shape represented by the subset of points. The system may then use the collider in determining whether a collision element collides with the particular feature.