Asymmetric Deep Supervised Hashing

Authors: Qing-Yuan Jiang, Wu-Jun Li

AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on three large-scale datasets show that ADSH can achieve state-of-the-art performance in real applications.
Researcher Affiliation Academia Qing-Yuan Jiang, Wu-Jun Li National Key Laboratory for Novel Software Technology Collaborative Innovation Center of Novel Software Technology and Industrialization Department of Computer Science and Technology, Nanjing University, China jiangqy@lamda.nju.edu.cn,liwujun@nju.edu.cn
Pseudocode Yes Algorithm 1 The learning algorithm for ADSH
Open Source Code No The paper mentions that source code for some baselines is provided by their authors, but there is no statement or link indicating that the authors of ADSH are providing their code.
Open Datasets Yes We evaluate ADSH on three datasets: MS-COCO (Lin et al. 2014b), CIFAR-10 (Krizhevsky 2009) and NUS-WIDE (Chua et al. 2009).
Dataset Splits Yes tune the learning rate among [10-6, 10-2] by using a validation set. For ADSH method, we set γ = 200, Tout = 50, Tin = 3, |Ω| = 2000 by using a validation strategy for all datasets.
Hardware Specification Yes We carry out experiments to evaluate our ADSH and baselines which are implemented with the deep learning toolbox Mat Conv Net (Vedaldi and Lenc 2015) on a NVIDIA M40 GPU server.
Software Dependencies No The paper mentions 'deep learning toolbox Mat Conv Net (Vedaldi and Lenc 2015)' but does not provide specific version numbers for software dependencies.
Experiment Setup Yes In order to avoid overfitting, we set weight decay as 5e-4. Following the suggestions of the authors, we set the mini-batch size to be 128 and tune the learning rate among [10-6, 10-2] by using a validation set. For ADSH method, we set γ = 200, Tout = 50, Tin = 3, |Ω| = 2000 by using a validation strategy for all datasets.