Learning to Hash on Partial Multi-Modal Data
Authors: Qifan Wang, Luo Si, Bin Shen
IJCAI 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on two multi-modal datasets demonstrate the superior performance of the proposed approach over several state-of-the-art multi-modal hashing methods. 3 Experimental Results We evaluate our method on two image datasets: NUS-WIDE and MIRFLICKR-25k. |
| Researcher Affiliation | Academia | Qifan Wang, Luo Si and Bin Shen Computer Science Department, Purdue University West Lafayette, IN 47907, US wang868@purdue.edu, lsi@purdue.edu, bshen@purdue.edu |
| Pseudocode | Yes | Algorithm 1 Partial Multi-Modal Hashing (PM2H) |
| Open Source Code | No | The paper does not explicitly state that the source code for their proposed method (PM2H) is publicly available or provide a link to it. |
| Open Datasets | Yes | NUS-WIDE1 contains 270k images associated with more than 5k unique tags. 81 ground-truth concepts are annotated on these images. We filter out those images with less than 10 tags, resulting in a subset of 110k image examples. Visual features are represented by 500-dimension SIFT [Lowe, 2004] histograms, and text features are represented by index vectors of the most common 2k tags. We use 90% of the data as the training set and the rest 10% as the query set. MIRFLICKR-25k2 is collected from Flicker images for image retrieval tasks. This dataset contains 25k image examples associated with 38 unique labels. 100-dimensional SIFT descriptors and 512-dimensional GIST descriptors [Oliva and Torralba, 2001] are extracted from these images as the two modalities. We randomly choose 23k image examples as the training set and 2k for testing. |
| Dataset Splits | Yes | We use 90% of the data as the training set and the rest 10% as the query set. We randomly choose 23k image examples as the training set and 2k for testing. The parameters α, λ and γ are tuned by 5-fold cross validation on the training set. |
| Hardware Specification | Yes | We implement our algorithm using Matlab on a PC with Intel Duo Core i5-2400 CPU 3.1GHz and 8GB RAM. |
| Software Dependencies | No | The paper mentions "Matlab" but does not specify its version or any other software dependencies with version numbers. |
| Experiment Setup | Yes | The parameters α, λ and γ are tuned by 5-fold cross validation on the training set. We set the maximum number of iterations to 100. To remove any randomness caused by random selection of training set and random initialization, all of the results are averaged over 10 runs. We empirically choose 5 in our experiments. |