Large Scale Similarity Learning Using Similar Pairs for Person Verification

Authors: Yang Yang, Shengcai Liao, Zhen Lei, Stan Li

AAAI 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Results on the challenging datasets of face verification (LFW and Pub Fig) and person re-identification (VIPe R) show that our algorithm outperforms the state-of-the-art methods.
Researcher Affiliation Academia Yang Yang, Shengcai Liao, Zhen Lei, Stan Z. Li Center for Biometrics and Security Research & National Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences, Beijing, China
Pseudocode Yes Algorithm 1 Large Scale Similarity Learning Using Similar Pairs Input: A data set of N d-dimensional similar training pairs {(x1, y1), ..., (x N, y N)} after PCA. Output: A Rd d and B Rd d. 1: normalize each data with the l2-norm; 2: compute Σm S and Σe S by Eq. 10; 3: compute Σ by Eq. 11; 4: compute A and B by Eq. 12.
Open Source Code No The paper does not provide an explicit statement or link indicating that the source code for the methodology is openly available.
Open Datasets Yes We evaluate the performances of the proposed LSSL on three publicly available challenging datasets: Labeled Faces in the Wild (LFW) (Huang et al. 2007), Public Figures Face Database (Pub Fig) (Kumar et al. 2009) and Viewpoint Invariant Pedestrian Recognition (VIPe R) (Gray, Brennan, and Tao 2007).
Dataset Splits Yes The images are divided into 10 folds that are used for cross-validation and there are 300 similar pairs and 300 dissimilar pairs in each fold. ... Image pairs of 9 folds are used for training while the remaining fold is used for testing. The average result over 10 trials is reported.
Hardware Specification Yes All of them are evaluated on a PC with the 3.40 GHz Core I7 CPU with 8 cores.
Software Dependencies No The paper does not provide specific version numbers for any software dependencies or libraries used in the experiments.
Experiment Setup Yes In experiments, we set λ to 0.9, 1.2 and 1.5 for LFW, Pub Fig and VIPe R, respectively.