3D Part Segmentation

3D part segmentation assigns points, mesh faces, or other 3D elements to components such as a chair leg, door handle, or movable object part, supporting editing, recognition, and robotic interaction. Methods combine geometric features and point-cloud networks with multi-view image cues, generative 3D priors, and motion-based reasoning; language and vision-language models increasingly enable promptable or open-vocabulary labels. Benchmarks such as ShapeNetPart and PartNet test part labeling and decomposition, while articulated-object datasets evaluate whether systems recover parts and their joints. A central challenge is producing consistent, useful part boundaries across unfamiliar categories, partial or noisy observations, and differing levels of detail, often with limited labeled data.