Unsupervised Learning of Multi-Level Descriptors for Person Re-Identification
Authors: Yang Yang, Longyin Wen, Siwei Lyu, Stan Li
AAAI 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on the challenging person re-identification datasets VIPe R and CUHK 01, demonstrate the effectiveness of our method. |
| Researcher Affiliation | Academia | 1Center for Biometrics and Security Research & National Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences, Beijing, China, 100190 2University at Albany, State University of New York, Albany, NY, 12222 |
| Pseudocode | No | No clearly labeled pseudocode or algorithm block was found in the paper. |
| Open Source Code | Yes | We have released the MATLAB code 1 for future research on person re-identification. 1http://www.cbsr.ia.ac.cn/users/yyang/main.htm. |
| Open Datasets | Yes | VIPe R Dataset. The viewpoint invariant pedestrian recognition (VIPe R) dataset contains 632 image pairs, corresponding to 632 pedestrians. It was captured by two cameras in outdoor academic environments. This dataset is the most widely used in person re-identification. ... CUHK 01 Dataset. CUHK 01 dataset has 971 persons and each person has two images in each of two camera views. It was collected in a campus environment. ... The datasets are cited: (Gray, Brennan, and Tao 2007) for VIPeR and (Li, Zhao, and Wang 2012) for CUHK 01. |
| Dataset Splits | No | The training set is formed from 50% of randomly chosen image pairs and the remaining 50% image pairs are used for testing. |
| Hardware Specification | Yes | This is evaluated on a PC with the 3.40 GHz Core I7 CPU with 8 cores. |
| Software Dependencies | No | No specific software dependencies with version numbers (e.g., Python 3.8, PyTorch 1.9) were found. Only "MATLAB code" is mentioned without a version. |
| Experiment Setup | Yes | Unless otherwise specified, we empirically set the parameters in our model as follows: 1) In (1), λ1 = 0.001d2 and λ2 = 0.001d2 with d is the dimension of raw pixel data. 2) The numbers of bases are 350, 70, 60 for learning pixel-level, patch-level and image-level descriptors, respectively. 3) In the patch-level, we adopt 7 7 with a stride of 1. 4) When holistic image features are extracted, we use 10 horizontal stripes for images. 5) For all datasets, the dimension of each holistic image feature is reduced to 60 by PCA. |