FAMOS: Feed-Forward 3D Articulation Modeling from Sparse Observations

Forum topic · 小凯 · 2026-09-19

Summary

FAMOS is a feed-forward model that predicts movable-part segmentation and joint parameters of articulated 3D objects from a sparse, unordered set of partial point clouds. Unlike most prior methods that infer articulation from a single observation and rely heavily on learned category-level shape priors, FAMOS jointly reasons over multiple observations and naturally supports a variable number of inputs, including a single view. The authors introduce a Multi-state Articulation Transformer that alternates between state-wise and global attention to aggregate articulation cues across observations, along with an observed articulation span objective that supervises the range of motion each part exhibits in the input views. To overcome limited dataset scale and diversity, they also present a procedural data generator that synthesizes self-annotated assets during training. Experiments on PartNet-Mobility, ACD, and ArtiCraft-10K show that FAMOS consistently outperforms feed-forward and optimization-based baselines. Paper: arXiv 2609.20817.

Overview

Field: Computer Vision Authors: Kevin Qu, Tao Sun, Massimiliano Viola, Liyuan Zhu, Zhizhuo Zhou, Sayan Deb Sarkar, Konrad Schindler, Iro Armeni Published: 2026-09-17 arXiv: 2609.20817

Summary (translated)

Modeling articulated objects from sparse monocular views is challenging because each observation reveals only partial geometry and motion evidence. Most feed-forward methods infer articulation from a single observation and therefore rely heavily on learned category-level shape priors. FAMOS, a feed-forward model, predicts movable-part segmentation and joint parameters from a sparse, unordered set of partial point clouds. The model jointly reasons over multiple observations and naturally supports a variable number of inputs, including a single view.

To aggregate articulation cues across observations, the authors introduce a Multi-state Articulation Transformer with alternating state-wise and global attention. They further propose an observed articulation span objective that supervises the range of motion each part exhibits across the input observations, encouraging the model to exploit the full observation set.

To overcome the limited scale and diversity of existing datasets, they introduce a procedural data generator that synthesizes self-annotated assets during training. Experiments on PartNet-Mobility, ACD, and ArtiCraft-10K show that FAMOS consistently outperforms both feed-forward and optimization-based baselines.

Project page: https://kevinqu7.github.io/famos

Key contributions

Feed-forward articulation prediction from sparse, unordered sets of partial point clouds, supporting a variable number of observations
Multi-state Articulation Transformer with alternating state-wise and global attention
Observed articulation span objective for cross-observation supervision
Procedural data generator producing self-annotated training assets
State-of-the-art results on PartNet-Mobility, ACD, and ArtiCraft-10K