SOML: Sparse Online Metric Learning with Application to Image Retrieval
Authors: Xingyu Gao, Steven C.H. Hoi, Yongdong Zhang, Ji Wan, Jintao Li
AAAI 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In our experiments, we investigate the application of the proposed sparse online metric learning technique for improving the Bag-of-Words (Bo W) representation in image retrieval tasks. In the following, we first introduce the experimental testbed and setup, followed by discussing the detailed experimental results. Our experimental results show that the proposed method achieves better or at least comparable accuracy performance than the state-of-the-art non-sparse distance metric learning approaches, but enjoys a significant advantage in computational efficiency and sparsity, making it more practical for real-world applications. |
| Researcher Affiliation | Academia | Xingyu Gao1,2,3, Steven C.H. Hoi2, Yongdong Zhang1, Ji Wan1,2,3, Jintao Li1 1Key Laboratory of Intelligent Information Processing of Chinese Academy of Sciences (CAS), Institute of Computing Technology, CAS, Beijing, China 2School of Computer Engineering, Nanyang Technological University, Singapore 3University of Chinese Academy of Sciences, Beijing, China |
| Pseudocode | Yes | Algorithm 1 SOML-TG Sparse Online Metric Learning via Truncated Gradient |
| Open Source Code | No | The paper does not contain any statement about making its source code publicly available, nor does it provide a link to a code repository for the described methodology. |
| Open Datasets | Yes | Following previous studies, we adopt the Oxford5K image dataset, a well-known public dataset for image retrieval benchmarks. This dataset contains a total of 5,062 images for 11 Oxford landmarks with manually annotated ground truth. |
| Dataset Splits | No | The paper describes how training triplet instances are generated (21,000 from 7 randomly selected positive images and 500 negative images) and states 'The remaining 4,555 images are used for testing/retrieval.' However, it does not specify a separate 'validation' set or explicit train/validation/test splits in percentages or absolute counts for the overall dataset used in model development beyond what's implicitly used for triplet generation and final evaluation. |
| Hardware Specification | No | The paper does not provide any specific details regarding the hardware used for running experiments, such as CPU or GPU models, memory, or specific computing environments. |
| Software Dependencies | No | The paper mentions software components and techniques like 'SIFT for feature descriptors', 'Approximate Kmeans (AKM) clustering', and refers to other algorithms (QPAO, OASIS), but does not provide specific version numbers for any software dependencies. |
| Experiment Setup | Yes | For parameter settings, we set the parameters for the three different-sized codebooks (10,000, 100,000, and 1-million) are: L = 0, M = 104, 50 subsets; L = 0, M = 105, 50 subsets; and L = 0, M = 106, 500 subsets for QPAO (Cai, Yan, and Mikolajczyk 2010). For OASIS, the parameters are C = 0.1 and 105 training steps with different-sized codebooks. For SOML-TG algorithm, we set parameters η = 1, λ = 10-5 for all the codebooks respectively. For SOMLDA algorithm, we set parameters γ = 10-4, ρ = 1, and λ = 10-6 with the three different-sized codebooks respectively. |