Semi-Supervised Bayesian Attribute Learning for Person Re-Identification
Authors: Wenhe Liu, Xiaojun Chang, Ling Chen, Yi Yang
AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments demonstrate the superior performance of our algorithm over several state-of-the-art algorithms on small-scale datasets and comparable performance on large-scale re-ID datasets. |
| Researcher Affiliation | Academia | 1Centre for Artificial Intelligence, University of Technology Sydney, Sydney, Australia. 2Language Technologies Institute, Carnegie Mellon University, Pittsburgh, PA, USA. |
| Pseudocode | Yes | Algorithm 1 Semi-supervised Bayesian Attribute Learning 1: Initialize K+ = 1,A = [i xi/N]. 2: while objective value in (18) deceasing do 3: for n = 1, , N do 4: for n = 1, , K+ do 5: Determine zij {0, 1} to minimize the objective value in (18) greedily; 6: end for 7: end for 8: A XZT (ZZT ) 1. 9: Sample a new basis a K+ with probability P(a K+ = xi Azi) xi Azi 2 2. 10: update A [A, a K+]; 11: update K+ K+ + 1. 12: update Θ which is the expectation of W as in (17) 13: update y . 14: end while |
| Open Source Code | No | For large-scale datasets, the numbers of their pair-wise labels are huge, we use a Stochastic Gradient SVM package Svm Sgd : http://leon.bottou.org/projects/sgd. |
| Open Datasets | Yes | The VIPe R dataset (Gray, Brennan, and Tao 2007) collects 1,264 images of 632 people from two nonoverlapping camera views. The PRID dataset (Hirzer et al. 2011) contains images of individuals from two distinct cameras. The Duke MTMC-re ID dataset (Zheng, Zheng, and Yang 2017) is a subset of the Duke MTMC dataset. |
| Dataset Splits | No | We randomly select 316 people as the testing set for the experiment; the ramaining people were used as the training set. |
| Hardware Specification | No | Following the pre-processing procedure outlined in (Lin et al. 2017), all images were first rescaled to 224 224 pixels. Then, we extracted 2048 dimensional feature vectors from the images using a pre-trained Res Net-50 deep neural network (He et al. 2016). |
| Software Dependencies | No | Then, we extracted 2048 dimensional feature vectors from the images using a pre-trained Res Net-50 deep neural network (He et al. 2016). For large-scale datasets, the numbers of their pair-wise labels are huge, we use a Stochastic Gradient SVM package Svm Sgd : http://leon.bottou.org/projects/sgd. Eq. (17) can be efficiently solved as a standard binary SVM problem with a vectorized matrix Z and Θ (Pirsiavash, Ramanan, and Fowlkes 2009) using public SVM solvers. |
| Experiment Setup | No | Following the pre-processing procedure outlined in (Lin et al. 2017), all images were first rescaled to 224 224 pixels. Then, we extracted 2048 dimensional feature vectors from the images using a pre-trained Res Net-50 deep neural network (He et al. 2016). We conducted experiments over ten splits and report the average results. |