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..
Diffusion-Driven Progressive Target Manipulation for Source-Free Domain Adaptation
Authors: Yuyang Huang, Yabo Chen, Junyu Zhou, Wenrui Dai, XIAOPENG ZHANG, Junni Zou, Hongkai Xiong, Qi Tian
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results demonstrate that DPTM outperforms existing methods by a large margin and achieves state-of-the-art performance on four prevailing SFDA benchmark datasets with different scales. Remarkably, DPTM can significantly enhance the performance by up to 18.6% in scenarios with large source-target gaps. |
| Researcher Affiliation | Collaboration | 1Shanghai Jiao Tong University, Shanghai, China 2Huawei Inc., Shenzhen, China |
| Pseudocode | No | The paper describes the method and framework in Section 3 and Figure 1, but no explicit pseudocode or algorithm block is present. |
| Open Source Code | Yes | All the code will be made public after acceptance. |
| Open Datasets | Yes | Datasets. We adopt four standard domain adaptation benchmarks of different scales for evaluations, including the small-scale Office-31 dataset [33], the medium-scale Office-Home dataset [41], and two large-scale datasets (i.e., Vis DA [28] and Domain Net-126 [27]). |
| Dataset Splits | No | Refer to the supplementary material for complete dataset statistics and domain configurations. |
| Hardware Specification | Yes | The computing requirements are provided in the Appendix. |
| Software Dependencies | No | We employ stable-diffusion v1-5 [32] as the diffusion model to generate 512 512 images with 20 denoising steps. While 'stable-diffusion v1-5' is a specific software component, the paper does not list multiple key software components with their specific version numbers (e.g., Python, PyTorch, CUDA versions). |
| Experiment Setup | Yes | We employ stable-diffusion v1-5 [32] as the diffusion model to generate 512 512 images with 20 denoising steps. γ1 = 5.5 in (1) and γ2 = 0 in (6). We set the threshold E to 0.01, and the total refinement iteration count R to 10. ... We train for 20K iterations with the batch size of 128 and learning rate of 3e-3 for large-scale Domain Net-126 [27] and Vis DA [28], and 15K iterations with the batch size of 32 and learning rate of 1e-3 for Office-31 and Office-Home. Weight decay is set to 5e-4 for all the datasets. |