Near-Optimal Joint Object Matching via Convex Relaxation

Authors: Yuxin Chen, Leonidas Guibas, Qixing Huang

ICML 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we evaluate the performance of Match Lift and compare it against (Cand es et al., 2011; Jalali et al., 2011; Jalali & Srebro, 2012) and two other graph matching methods. We consider both synthetic examples, which are used to verify the exact recovery conditions described above, as well as popular benchmark datasets for evaluating the practicability on real-world images.
Researcher Affiliation Academia Yuxin Chen YXCHEN@STANFORD.EDU Department of Electrical Engineering, Stanford University, Stanford, CA 94305, USA Leonidas Guibas GUIBAS@CS.STANFORD.EDU Department of Computer Science, Stanford University, Stanford, CA 94305, USA Qixing Huang HUANGQX@STANFORD.EDU Department of Computer Science, Stanford University, Stanford, CA 94305, USA
Pseudocode Yes Algorithm 1 Estimating the size m of the universe S
Open Source Code No The paper mentions that the algorithm is detailed in (Chen et al., 2014) and the supplemental materials, but does not provide an explicit statement or link for open-source code for the methodology described in this paper.
Open Datasets Yes We have applied our algorithm on six benchmark datasets, i.e., CMU-House, CMU-Hotel, two datasets (Graf and Bikes) from (Mikolajczyk & Schmid, 2005) and two new datasets (referred as Chair and Building, respectively) designed for evaluating joint partial object matching.
Dataset Splits No The paper mentions using benchmark datasets and generating synthetic examples, but does not specify train/validation/test dataset splits (e.g., percentages or sample counts) needed to reproduce the experiments.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory amounts, or detailed computer specifications) used for running its experiments.
Software Dependencies No The paper mentions various algorithms and solvers (e.g., 'Se Du Mi or MOSEK', 'ADMM method') and techniques (e.g., 'RANSAC', 'SIFT') but does not provide specific version numbers for any software dependencies.
Experiment Setup Yes For simplicity, we only consider the full observation mode, which establishes input maps between all pairs of objects. In all examples, we fix the universe size such that it consists of m = 16 points. We then vary the remaining parameters, i.e., n, pset and pfalse, to assess the performance of an algorithm. We evaluate 31 x 36 sets of parameters for each scenario, where each parameter configuration is simulated by 10 Monte Carlo trials.