Search papers, labs, and topics across Lattice.
This paper introduces Orbit-Space Geometric Probability Paths (OGPP), a particle-native flow-matching framework designed for generative modeling of particle systems by explicitly addressing permutation symmetries and encoding geometric attributes into the flow's terminal velocity. OGPP uses orbit-space canonicalization, particle index embeddings, and geometric probability paths with arc-length-aware terminal velocities to improve performance. Experiments on minimal-surface benchmarks, ShapeNet, and single-shape encoding demonstrate significant improvements in metric error, sample quality, and computational efficiency compared to existing methods.
By explicitly modeling permutation symmetries and geometric attributes, OGPP achieves state-of-the-art generative performance on particle systems with significantly fewer steps and parameters.
We present Orbit-Space Geometric Probability Paths (OGPP), a particle-native flow-matching framework for generative modeling of particle systems. OGPP is motivated by two insights: (i) particles are defined up to permutation symmetries, so anonymous indexing inflates per-index target variance and yields curved, hard-to-learn flows; and (ii) particles live in physical space, so the flow terminal velocity has physical meaning and can encode geometric attributes, e.g., surface normals. OGPP instantiates three key components: (1) orbit-space canonicalization of the probability-path terminal endpoint, (2) particle index embeddings for role specialization, and (3) geometric probability paths with arc-length-aware terminal velocities that generate normals as a byproduct of the flow. We evaluate OGPP on minimal-surface benchmarks, where it reduces metric error by up to two orders of magnitude in a single inference step; on ShapeNet, where it matches the state of the art with 5x fewer steps and reaches airplane EMD comparable to DiT-3D with 26x fewer parameters and 5x fewer steps; and on single-shape encoding, where it produces normals and reconstructions competitive with 6D generators while operating entirely in 3D.
