Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Dynamic Masking and Auxiliary Hash Learning for Enhanced Cross-Modal Retrieval
Authors: Shuang Zhang, Yue Wu, Lei Shi, Yingxue Zhang, Feifei Kou, Huilong Jin, Pengfei Zhang, Meiyu Liang, Mingying Xu
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Comprehensive experimental results on NUS-WIDE, MIRFlickr-25K and MS-COCO benchmark datasets show that the proposed AHLR algorithm outperforms several existing algorithms. |
| Researcher Affiliation | Academia | Shuang Zhang1,2, Yue Wu1, Lei Shi3, , Yingxue Zhang4, Feifei Kou5,6, Huilong Jin1, Pengfei Zhang7, Meiyu Liang5, Mingying Xu8 1College of Engineering, Hebei Normal University 2Hebei Provincial Key Laboratory of Information Fusion and Intelligent Control 3State Key Laboratory of Media Convergence and Communication, Communication University of China 4College of Computer and Cyber Security, Hebei Normal University 5School of Computer Science (National Pilot School of Software Engineering), BUPT 6key Laboratory of Trustworthy Distributed Computing and Service, BUPT, Ministry of Education 7School of Computer Science and Engineering, Anhui University of Science of Technology 8School of Artificial Intelligence and Computer Science, North China University of Technology *Corresponding author: EMAIL |
| Pseudocode | No | The paper describes the methods in text and mathematical formulas but does not include any explicit pseudocode blocks or algorithms. |
| Open Source Code | Yes | The source code associated with our paper can be accessed on Git Hub at the following link: https://anonymous.4open.science/r/AHLR-C48C. |
| Open Datasets | Yes | The three datasets used in this study and their download links are as follows: MIRFlickr25K:https://www.kaggle.com/datasets/paulrohan2020/ mirflickr25k; NUS-WIDE:https://lms.comp.nus.edu.sg/wp-content/uploads/2019/ research/nuswide/NUS-WIDE.html; MS-COCO: https://cocodataset.org/#download. |
| Dataset Splits | Yes | MIRFlickr-25K contains 25,000 images... We select image-text pairs with at least 20 labels as experimental data, randomly extract 2,000 pairs as query sets, the rest as retrieval sets, and randomly select 10,000 pairs from the retrieval sets as training sets. NUS-WIDE contains 269,648 web images... We select 186,577 image-text pairs, all of which belong to the 10 most common classes, randomly select 2,100 pairs from the dataset as query sets, the rest as retrieval sets, and randomly select 10,500 pairs from the retrieval sets as training sets. MS-COCO contains 123,289 images... 5,000 pairs were randomly selected from the dataset as the query set, the rest as the retrieval set, and 10,000 pairs were randomly selected from the retrieval set as training data. |
| Hardware Specification | Yes | The AHLR method is mainly implemented based on Pytorch[16], and all experiments are run on a server equipped with an NVIDIA Ge Force RTX 3080 graphics card with 40GB RAM to ensure the stability of the experiment. |
| Software Dependencies | Yes | The AHLR method is mainly implemented based on Pytorch[16]... |
| Experiment Setup | Yes | In the experiment, the batch size is set to 64, the Adam optimizer[15] is used for the main optimization, and the method of dynamically adjusting the learning rate is adopted, where the initial learning rate of 1e-3, a decay schedule of 0.9 times the learning rate every 5 epochs, and a weight decay of 0.2. |