Robust Model Reasoning and Fitting via Dual Sparsity Pursuit

Authors: Xingyu Jiang, Jiayi Ma

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

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments regarding known and unknown model fitting on synthetic and challenging real datasets have demonstrated the superiority of our method against the stateof-the-art.
Researcher Affiliation Academia 1Huazhong University of Science and Technology, 2Wuhan University
Pseudocode Yes We conclude the pseudo code of the implementation of our DSP method in Alg. 1 and Alg. 2.
Open Source Code Yes Code is released at: https://github.com/Sta Rain J/DSP.
Open Datasets Yes 8 public datasets [8, 2] are used, and we divide them into two groups including Fund: kusvod2, CPC, TUM, KITTI, T&T; and Homo: homogr, EVD, Hpatch.
Dataset Splits No The paper describes the synthetic data generation parameters and mentions the use of public datasets for testing, but it does not specify explicit training/validation/test splits for their method.
Hardware Specification Yes The experiments of RANSAC [19], EAS [18] 2 and our DSP are conducted on a desktop with 4.0 GHz Intel Core i7-6700K CPU and 16GB memory. ... And two deep learning methods are accelerated by NVIDIA TITAN V GPUs.
Software Dependencies No The paper mentions 'MATLAB code' and 'Ubuntu 16.04' but does not specify version numbers for other ancillary software libraries or solvers used in their method.
Experiment Setup Yes In DSP, λ and γ are two hyper-parameters. Based on [44], we set λ = 0.005 log(4N) [1, 1, 0.5, 1, 1, 0.5, 0.5, 0.5, 0.1] as default... In addition, we set γ = 0.06 at the beginning, then update it with 0.98γ for each twenty iterations, and constrain γmin = 0.02. Moreover, we set the max iteration as 2k, and stop it if ε = xk xk 1 2 1e 6. As for τ, it controls the number of estimated basis, i.e., r. We set ξ = L(M, xi, ei)/L(M, xi 1, ei 1), and describe its distribution on all real data as in Fig. 3. Based on the best ξ, we set τ = 1.2L(M, xi 1, ei 1) during the estimation of xi.