Online Cross-Modal Hashing for Web Image Retrieval
Authors: Liang Xie, Jialie Shen, Lei Zhu
AAAI 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results on two real-world multi-modal web image datasets: MIR Flickr and NUS-WIDE, demonstrate the effectiveness and efficiency of OCMH for online cross-modal web image retrieval. |
| Researcher Affiliation | Academia | Liang Xie Department of Mathematics Wuhan University of Technology, China whutxl@hotmail.com Jialie Shen and Lei Zhu School of Information Systems Singapore Management University, Singapore jlshen@smu.edu.sg, leizhu0608@gmail.com |
| Pseudocode | Yes | Algorithm 1 Optimizing algorithm at round t |
| Open Source Code | No | The paper does not provide any explicit statement or link for open-source code for the described methodology. |
| Open Datasets | Yes | In our experiments, two real-world web image datasets: MIR Flickr and NUS-WIDE, are used to evaluate the effectiveness and efficiency of OCMH. ... On MIR Flickr, we directly use the image and text features provided in (Guillaumin, Verbeek, and Schmid 2010), including 15 image features and one binary text feature. We also directly use 6 image features and one binary text feature provided by (Chua et al. 2009). |
| Dataset Splits | Yes | To support the evaluation of online performance, the whole dataset is split to 13 data chunks, each of the first 12 chunks contains 2,000 pairs, and the last chunk contains 1,000 pairs. NUS-WIDE contains 269,648 image-text pairs which are also downloaded from Flickr, each pair is labeled by 81 concepts that can be used for evaluation. ... The whole dataset is split to 27 chunks, each of the first 26 chunks contains 10,000 pairs, and the last chunk contains 9,648 pairs. |
| Hardware Specification | Yes | All the experiments are conducted on a computer with Intel Core(TM) i5 2.6GHz 2 processors and 12.0GB RAM. |
| 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 the implementation of OCMH, we set the regularization parameters λ, α and β to 10 6. Since text usually contains more semantic information than image, we set θ1 = 0.3 and θ2 = 0.7. ... At each round, since Vm has been optimized by old data, we do not need much iterations for updating Vm, as well as the updating of H. Thus we set Titer = 3 in our implementations. |