Researchers have introduced FAMOS, a feed-forward model for 3D articulation modeling that works from sparse monocular observations. The work addresses a core difficulty: a single view of an articulated object reveals only partial geometry and motion evidence, which forces most existing feed-forward methods to lean heavily on learned category-level shape priors.
FAMOS instead operates on a sparse, unordered set of partial point clouds. It jointly reasons over multiple observations and supports a variable number of inputs, including a single view, allowing articulation prediction to be grounded in observed motion rather than shape priors alone.
To aggregate articulation cues across observations, the authors propose a Multi-state Articulation Transformer with alternating state-wise and global attention. They also introduce an observed articulation span objective that supervises the motion range each part exhibits across the input observations, encouraging the model to use the full observation set.
Because existing datasets are limited in scale and diversity, the team built a procedural data generator that synthesizes self-annotated assets during training. Experiments on PartNet-Mobility, ACD, and ArtiCraft-10K reportedly show consistent improvements over both feed-forward and optimization-based baselines.
The project page is available at kevinqu7.github.io/famos.