Locality Preserving Refinement for Shape Matching with Functional Maps

Authors: Yifan Xia, Yifan Lu, Yuan Gao, Jiayi Ma

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on public benchmarks demonstrate the superiority of our method over the state-of-the-art methods. In this section, we apply our LOPR to challenging shapes from several public datasets and compare it with classical and state-of-the-art approaches.
Researcher Affiliation Academia Yifan Xia, Yifan Lu, Yuan Gao , Jiayi Ma* Electronic Information School, Wuhan University, Wuhan 430072, China xiayifan@whu.edu.cn, lyf048@whu.edu.cn, ethan.y.gao@gmail.com, jyma2010@gmail.com
Pseudocode Yes Algorithm 1: Locality Preserving Refinement for Pointwise Map Recovery
Open Source Code Yes Our code is publicly available at https://github.com/Xia Yifan1999/LOPR.
Open Datasets Yes Datasets Four public datasets are used in our exvluation experiments: FAUST (Bogo et al. 2014) contains a total of 100 shapes, representing 10 poses of 10 different human subjects... TOSCA (Bronstein, Bronstein, and Kimmel 2008) provides 76 shapes divided into 8 distinct categories... SCAPE (Anguelov et al. 2005) contains 71 registered meshes... TOPKIDS (L ahner et al. 2016) consists of 26 nonintersecting manifold shapes...
Dataset Splits No The paper mentions selecting shape pairs for quantitative evaluation (e.g., 'randomly selected 300 shape pairs' for FAUST), but it does not specify explicit training, validation, and test splits or their proportions/counts.
Hardware Specification Yes All experiments are conducted on a PC with Intel(R) Core i9-9920X CPU at 3.50GHz, using MATLAB R2018a. And K-nearest neighbor search is accelerated by GPU.
Software Dependencies Yes All experiments are conducted on a PC with Intel(R) Core i9-9920X CPU at 3.50GHz, using MATLAB R2018a.
Experiment Setup Yes Parameter λ is the crucial threshold to determine the inlier set I. To find an appropriate value, we select 200 shape pairs from the FAUST dataset... Fig. 3 shows that λ can be set to 0.2... K1 = |I|/100 , where means rounding up. Similarly, as the number of nearest neighbors searched in the spectral domain in Eq. (17), K2 also deserves to be proportional to |I|, e.g., K2 = |I|/1000 . Here, the functional map matrices are uniformly computed utilizing MWP with 5 iterations.