Image Cosegmentation via Saliency-Guided Constrained Clustering with Cosine Similarity

Authors: Zhiqiang Tao, Hongfu Liu, Huazhu Fu, Yun Fu

AAAI 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results on two widely-used datasets demonstrate our approach achieves competitive performance over the state-of-the-art cosegmentation methods.
Researcher Affiliation Academia 1Department of Electrical and Computer Engineering, Northeastern University, Boston, USA, 02115. 2Institute for Infocomm Research, Agency for Science, Technology and Research, Singapore. 3College of Computer and Information Science, Northeastern University, Boston, USA, 02115.
Pseudocode Yes Algorithm 1 The algorithm of SGC3 for cosegmentation
Open Source Code No The paper does not provide a link or explicit statement about the availability of its source code.
Open Datasets Yes We conduct our experiment on two benchmark datasets, which are i Coseg dataset2 (Batra et al. 2011), and Internet dataset3 (Rubinstein et al. 2013), respectively. 2http://chenlab.ece.cornell.edu/projects/touch-coseg/ 3http://people.csail.mit.edu/mrub/Object Discovery/
Dataset Splits No The paper describes using specific subsets of public datasets for evaluation (e.g., 'selects 31 image groups with 530 images' and '100 images per class'). However, it does not specify explicit training, validation, or test dataset splits in the typical supervised learning context (e.g., 80/10/10 percentages or separate subsets for these distinct phases).
Hardware Specification Yes All the experiments were conducted by MATLAB on a 64-bit Windows platform with two Intel Core i7 3.4GHz CPUs and 32GB RAM.
Software Dependencies No The paper mentions 'MATLAB' but does not provide a specific version number or version numbers for any other software libraries or dependencies used.
Experiment Setup Yes The weights of these three descriptors are set to be 0.6, 0.2 and 0.2, respectively. Moreover, we set the λ in Eq. (2) as 1e3 as the default setting. ...for each individual descriptor, we obtained 300 words by perform K-means clustering on each image group with the superpixel-level feature.