Cross-View Feature Learning for Scalable Social Image Analysis

Authors: Wenxuan Xie, Yuxin Peng, Jianguo Xiao

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

Reproducibility Variable Result LLM Response
Research Type Experimental In Section 6, the performance of our proposed method is evaluated on two real-world datasets in image classification.
Researcher Affiliation Academia Wenxuan Xie and Yuxin Peng and Jianguo Xiao Institute of Computer Science and Technology, Peking University, Beijing 100871, China
Pseudocode No The paper describes the CVFL framework and its solution mathematically but does not contain any structured pseudocode or algorithm blocks.
Open Source Code No The paper mentions running MATLAB codes but does not provide concrete access to source code for the methodology described, nor does it state that the code is publicly available.
Open Datasets Yes We conduct experiments on two publicly-available datasets in the multi-class classification setting. The first dataset is Corel-5K (Corel for short) (Duygulu et al. 2002), which is composed of 50 categories... The second dataset is NUS-WIDE-Object (NUS for short) (Chua et al. 2009), which is collected from the photo sharing website Flickr.
Dataset Splits Yes Following the partition of the dataset, 4,500 images are used for training and the rest are used for test. ...we follow the standard train/test partition of the dataset, where the training set and the test set contain 14,270 and 9,683 images, respectively. ...Parameters are determined by 5-fold cross-validation in the experiments.
Hardware Specification Yes Note that we run MATLAB codes on a server with 2.20GHz CPU and 128GB RAM.
Software Dependencies No The paper mentions 'MATLAB codes' and 'linear SVM (Fan et al. 2008)' but does not provide specific version numbers for any software dependencies.
Experiment Setup Yes Parameters are determined by 5-fold cross-validation in the experiments. We empirically find that the performance of CVFL reaches its peak when both λ and γ are chosen between 1 and 2. Concretely, we choose λ = 1.5 and γ = 1.8 for the Corel dataset, and choose λ = 1.7 and γ = 1.9 for the NUS dataset. ...we adopt linear SVM (Fan et al. 2008) to obtain the classification accuracy for evaluation.