UniPixie: Unified and Probabilistic 3D Physics Learning via Flow Matching

2026-06-03

AI Summary

Existing feedforward networks are limited to predicting a single physical property and struggle to capture the inherent ambiguity and diversity of material characteristics. This work proposes a unified neural architecture that reframes physical property prediction as the learning of controllable, continuous material attribute distributions, enabling the generation of a continuous parameter trajectory—from softest to hardest—from a single input image. Built upon flow matching and trained on the PIXIEMULTIVERSE dataset, the method produces simulation-ready parameters compatible with multiple physics solvers, including MPM, LBS, and mass-spring systems. Experiments demonstrate a reduction of over 50% in Young’s modulus prediction error compared to the strongest deterministic baseline, while generating rich, physically plausible dynamic behaviors. To our knowledge, this is the first approach to enable controllable, continuous material field generation across diverse physics engines.

Abstract

Existing feed-forward networks excel at predicting a single set of physical properties from visual appearance, but this point-estimate paradigm fundamentally fails to capture the real world's inherent physical ambiguity. We address this by reframing physics prediction as a task of learning a controllable, continuous distribution of material properties. We introduce UNIPIXIE, a framework trained to predict a continuous and parameterized path of physically plausible material properties from a single visual input. By learning a direct mapping along an object's softest-to-stiffest spectrum on our PIXIEMULTIVERSE dataset, UNIPIXIE allows for controllable generation of diverse, physically valid material fields via a single intuitive parameter. Crucially, UNIPIXIE introduces a novel unified architecture to produce simulation-ready parameters for diverse physics solvers, including continuum-based Material Point Method (MPM), reduced-order deformation based on Linear Blend Skinning (LBS), and anchor-based Spring-Mass systems, addressing a key portability issue in prior work. Experiments show our approach not only generates a rich variety of plausible dynamics but also reduces Young's Modulus prediction error by over 50% against the strongest deterministic baseline, bridging the gap between static point estimates and the continuous nature of physical reality. Project page: https://unipixie.github.io/

Problem

• physical ambiguity
• material property distribution
• physics prediction
• parameter portability
• continuous physical reality

Innovation

• flow matching
• probabilistic physics prediction
• unified material representation
• controllable physical simulation
• cross-solver portability