Deep Unified Cross-Modality Hashing by Pairwise Data Alignment

Authors: Yimu Wang, Bo Xue, Quan Cheng, Yuhui Chen, Lijun Zhang

IJCAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on three representative image-text datasets demonstrate the superiority of our DUCMH over several state-of-the-art cross-modality hashing methods.
Researcher Affiliation Academia National Key Laboratory for Novel Software Technology, Nanjing University, Nanjing 210023, China {wangym, xueb, chengq, chenyuhui, zhanglj}@lamda.nju.edu.cn
Pseudocode Yes Algorithm 1 The alternative learning algorithm
Open Source Code No No explicit statement or link providing access to open-source code was found.
Open Datasets Yes Three datasets, MIRFLICKR-25K [Huiskes and Lew, 2008], IAPR TC-12 [Escalante et al., 2010], and NUS-WIDE [Chua et al., 2009] are used for evaluation.
Dataset Splits No For MIRFLICKR-25K and IAPR TC-12, 2,000 data points are randomly sampled as the test (query) set, while for NUS-WIDE, 2,100 data points are selected. The remaining points as the retrieval set (database).
Hardware Specification Yes Our DUCMH method is implemented based on Py Torch [Paszke et al., 2019] with eight NVIDIA V100 GPUs and optimized by the mini-batch SGD with the size of 64 and weight decay.
Software Dependencies No The paper mentions 'Py Torch [Paszke et al., 2019]' but does not specify a version number for it or other software dependencies.
Experiment Setup Yes Our DUCMH method is implemented based on Py Torch [Paszke et al., 2019] with eight NVIDIA V100 GPUs and optimized by the mini-batch SGD with the size of 64 and weight decay. The learning rate is initialized as 0.0001 for the image to text mapping fi2t( ) and 0.004 for the unified hash function hy( ). Hyper-parameters ϵ, α and ρ are empirically set to 5000, 50 and 200 for scaling the order of each loss.