Dual Principal Component Pursuit: Improved Analysis and Efficient Algorithms

Authors: Zhihui Zhu, Yifan Wang, Daniel Robinson, Daniel Naiman, René Vidal, Manolis Tsakiris

NeurIPS 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on road plane detection from 3D point cloud data demonstrate that DPCP-PSGM can be more efficient than the traditional RANSAC algorithm, which is one of the most popular methods for such computer vision applications.
Researcher Affiliation Academia Zhihui Zhu MINDS Johns Hopkins University zzhu29@jhu.edu Yifan Wang SIST Shanghai Tech University wangyf@shanghaitech.edu.cn Daniel Robinson AMS Johns Hopkins University daniel.p.robinson@jhu.edu Daniel Naiman AMS Johns Hopkins University daniel.naiman@jhu.edu Rene Vidal MINDS Johns Hopkins University rvidal@jhu.edu Manolis C. Tsakiris SIST Shanghai Tech University mtsakiris@shanghaitech.edu.cn
Pseudocode Yes Algorithm 1 (DPCP-PSGM) Projected Sub-gradient Method for Solving (2) Input: data e X RD L and initial step size µ0; Initialization: set bb0 = arg minb e X b 2, s. t. b SD 1; 1: for k = 1, 2, . . . do 2: update the step size µk according to a certain rule; 3: bk = bbk 1 µk e X sign( e X bbk 1); bbk = PSD 1 (bk) = bk/ bk ; 4: end for
Open Source Code No No explicit statement providing access to the source code for the methodology described in this paper was found. No specific repository link was provided.
Open Datasets Yes Experiments on road plane detection from 3D point cloud data using the KITTI dataset [6], which is an important computer vision task in autonomous car driving systems
Dataset Splits No The paper states it manually annotated a few frames from the KITTI dataset, but does not provide specific details on train/validation/test splits (percentages, sample counts, or citations to predefined splits) needed for reproducibility.
Hardware Specification Yes Since DPCP-PSGM is the fastest method (on average converging in about 100 milliseconds for each frame on a 6 core 6 thread Intel (R) i5-8400 machine)
Software Dependencies No The paper mentions 'Gurobi [8]' as an efficient LP solver but does not provide a specific version number. No other key software components are listed with version numbers.
Experiment Setup Yes We set K0 = 30, K = 4 and β = 1/2 for the PGD step size with initial step size obtained by one iteration of a backtracking line search and denote the corresponding algorithm by PSGM-PGD. We define bb0 to be the bottom eigenvector of e X e X , which has been demonstrated to be effective in practice [24].