Geometry Processing with Neural Fields

Authors: Guandao Yang, Serge Belongie, Bharath Hariharan, Vladlen Koltun

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results show that our methods are on par with the well-established mesh-based methods without committing to a particular surface discretization.
Researcher Affiliation Collaboration Guandao Yang Cornell University Serge Belongie University of Copenhagen Bharath Hariharan Cornell University Vladlen Koltun Intel Labs
Pseudocode No The paper does not contain any figures, blocks, or sections explicitly labeled 'Pseudocode' or 'Algorithm'.
Open Source Code Yes Code is available at https://github.com/stevenygd/NFGP.
Open Datasets Yes We follow prior works [20, 76] to use Armadillo [40] and a sphere with one half of it corrupted by Gaussian noise. To create neural fields from these meshes, we follow the procedure of Park et al. [56] to compute ground-truth SDF for locations sampled within [ 1, 1]3.
Dataset Splits No The paper does not explicitly provide specific train/validation/test dataset splits (e.g., percentages, sample counts, or predefined split references) for its experiments.
Hardware Specification Yes Our deformation method right now requires a Titan X GPU with 12GB memory to train for 10 hours. Our smoothing and sharpening method takes about 10 minutes on the same GPU.
Software Dependencies No The paper mentions algorithms, optimizers (Adam), and methods (SIREN, Marching Cubes) but does not provide specific version numbers for any software dependencies.
Experiment Setup No The paper states 'Hyperparameters are provided in the supplement', but does not list specific hyperparameter values or detailed training configurations in the main text.