A Statistical Manifold Framework for Point Cloud Data

Authors: Yonghyeon Lee, Seungyeon Kim, Jinwon Choi, Frank Park

ICML 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments with large-scale standard benchmark point cloud data show greatly improved classification accuracy visa-vis existing methods.
Researcher Affiliation Collaboration 1Department of Mechanical Engineering, Seoul National University, Seoul, South Korea 2Kakao Enterprise, Seongnam, Kyonggi-do, South Korea 3Saige Research, Seoul, South Korea.
Pseudocode No The paper describes methods and formulations in text and mathematical equations, but does not include any dedicated pseudocode blocks or algorithm listings.
Open Source Code Yes Code is available at https://github.com/seungyeon-k/SMFpublic.
Open Datasets Yes Experiments are carried out with both synthetic and standard benchmark datasets (Shape Net (Chang et al., 2015), Model Net (Wu et al., 2015)).
Dataset Splits Yes In Section 4.1.1, we use a dataset consisting of cones, cylinders, and ellipsoids, which are split into training/validation/test sets of size 3196/800/804.
Hardware Specification No The paper does not provide specific hardware details such as GPU or CPU models used for experiments.
Software Dependencies No The paper mentions software components like 'ADAM' and 'DGCNN' and refers to their use, but does not specify version numbers for these or any other software dependencies like Python, PyTorch/TensorFlow, or CUDA.
Experiment Setup Yes To train the networks, we use ADAM with a learning rate of 0.001 and batch size of 16; the total number of the epochs is 500. The mean value of MEDs of the dataset is 0.0339, and we use the bandwidth value k to 0.5. We use Chamfer distance as the reconstruction loss; for regularization figures, the regularization term with the version of Equation (12) is multiplied by a coefficient λ = 107 with the info-Riemannian metric and by a coefficient λ = 1 with the Euclidean metric and added to the reconstruction loss term for each metric case. The value of η is set to be 0.0.