Efficient Graph Field Integrators Meet Point Clouds

Authors: Krzysztof Marcin Choromanski, Arijit Sehanobish, Han Lin, Yunfan Zhao, Eli Berger, Tetiana Parshakova, Alvin Pan, David Watkins, Tianyi Zhang, Valerii Likhosherstov, Somnath Basu Roy Chowdhury, Kumar Avinava Dubey, Deepali Jain, Tamas Sarlos, Snigdha Chaturvedi, Adrian Weller

ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We also perform exhaustive empirical evaluation, including on-surface interpolation for rigid and deformable objects (particularly for mesh-dynamics modeling), Wasserstein distance computations for point clouds, and the Gromov Wasserstein variant.
Researcher Affiliation Collaboration 1Google Research 2Columbia University 3Independent Researcher 4Haifa University 5Stanford University 6The Boston Dynamics AI Institute 7University of Cambridge 8The University of North Carolina at Chapel Hill 9The Alan Turing Institute.
Pseudocode Yes Algorithm 1 Fast Computation of Wasserstein Barycenter
Open Source Code Yes Our code is available at https://github.com/topographers/ efficient_graph_algorithms.
Open Datasets Yes We run tests on 120 meshes for 3D-printed objects with a wide range of sizes from the Thingi10k (Zhou & Jacobson, 2016) dataset (see Sec. C for details).
Dataset Splits Yes In each mesh, we randomly select a subset V V with |V | = 0.8|V| and mask out their vertex normals (set as zero vectors).
Hardware Specification Yes All experiments are run on a single computer with an i9-12900k CPU and 96GB memory.
Software Dependencies No For the baseline experiments, we use the implementation from the POT library (Flamary et al., 2021) for the GW-cg and FGW variants, and official implementation from (Xu et al., 2019) for the GW-prox variant.
Experiment Setup Yes For all our experiments, m = 16 random features, ϵ = 0.3, and the smoothing factor λ = 0.2 are chosen.