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..
Domain Adaptive Hashing Retrieval via VLM Assisted Pseudo-Labeling and Dual Space Adaptation
Authors: Jingyao Li, Zhanshan Li, Shuai Lü
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on three benchmark datasets demonstrate that VPDS consistently outperforms existing methods in both cross-domain and single-domain retrieval tasks, highlighting its effectiveness and superiority. [...] Experiments on three standard benchmark datasets demonstrate that VPDS achieves superior performance in both crossdomain and single-domain retrieval scenarios. [...] We evaluate our method on three benchmark datasets. |
| Researcher Affiliation | Academia | 1College of Computer Science and Technology, Jilin University, Changchun 130012, China 2Key Laboratory of Symbolic Computation and Knowledge Engineering (Jilin University), Ministry of Education, Changchun 130012, China Email addresses: EMAIL, EMAIL, EMAIL |
| Pseudocode | No | The paper describes its methodology through textual explanations and mathematical equations (e.g., Eq. 1, Eq. 2, Eq. 8, Eq. 9, Eq. 10, Eq. 11), but does not include any clearly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | The code is available in Supplementary, and the datasets are publicly available. |
| Open Datasets | Yes | Datasets. We evaluate our method on three benchmark datasets. Office-Home [48] consists of 4 domains from 65 classes. Consistent with prior works [3, 50, 51], we construct 6 tasks based on this dataset. Office-31 [37] contains 3 domains, with each domain containing 31 classes. Digits consists of 2 domains, MNIST [21] and USPS [16], and each domain contains 10 handwritten digits. [...] The code is available in Supplementary, and the datasets are publicly available. |
| Dataset Splits | Yes | For both cross-domain and single-domain retrieval tasks, 10% of the target domain samples are randomly selected as the query set. The remaining 90% along with all source domain samples constitute the training set. [...] To assess the generalization (out-of-sample) ability of our method, we split the source domain data into a 70% training set and a 30% retrieval set, and we divide the target domain data into a 60% training set, a 30% retrieval set and a 10% query set. |
| Hardware Specification | Yes | All experiments are conducted on NVIDIA A30 GPU. |
| Software Dependencies | No | The feature encoder F is implemented using the VGG [39] backbone. For the VLM, we adopt the pre-trained CLIP [36] with Vi T-B/16 [4], and keep its parameters frozen during training. The model is optimized using SGD with momentum 0.9 and weight decay 1e-5. |
| Experiment Setup | Yes | The model is optimized using SGD with momentum 0.9 and weight decay 1e-5. We set the initial learning rate to 1e-3 and use a batch size of 72. Training epoch is set to 20 for Digits and 100 for others. The hyperparameters are fixed as α = 0.9, η = 0.2, γ = 0.2 across all datasets. |