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..
Dual Data Alignment Makes AI-Generated Image Detector Easier Generalizable
Authors: Ruoxin Chen, Junwei Xi, Zhiyuan Yan, Ke-Yue Zhang, Shuang Wu, Jingyi Xie, Xu Chen, Lei Xu, Isabel Guan, Taiping Yao, Shouhong Ding
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our extensive evaluations demonstrate that a detector trained exclusively on DDA-aligned MSCOCO improves across diverse benchmarks. Code is available at https://github.com/roy-ch/Dual-Data-Alignment. ... Section 4 Experiments |
| Researcher Affiliation | Collaboration | Ruoxin Chen1, Junwei Xi2, Zhiyuan Yan3, Keyue Zhang1, Shuang Wu1, Jingyi Xie4, Xu Chen2, Lei Xu5, Isabel Guan6 , Taiping Yao1 , Shouhong Ding1 1Tencent You Tu Lab, 2East China University of Science and Technology, 3Peking University, 4Renmin University of China, 5Shenzhen University, 6Hong Kong University of Science and Technology |
| Pseudocode | No | The paper describes the DDA technique in prose: 'DDA consists of three steps: 1) VAE reconstruction for pixel alignment, 2) high-frequency fusion to eliminate bias, and 3) pixel mixup for further alignment in the pixel domain.' These steps are described in text, not in a structured pseudocode or algorithm block. |
| Open Source Code | Yes | Code is available at https://github.com/roy-ch/Dual-Data-Alignment. |
| Open Datasets | Yes | All compared detectors are evaluated on eleven diverse datasets, including seven benchmark datasets (Gen Image [59], DRCT-2M [4], Synthbuster [1], DDA-COCO, Eval GEN, AIGCDetection Benchmark [56] and Foren Synths [43] ) and four in-the-wild datasets (Chameleon [48], Wild RF [2], Synth Wildx [7] and BFree-Online [14]), where images are sourced from the web. |
| Dataset Splits | No | The training data exclusively consists of MSCOCO [29] images and their DDA-aligned counterparts. ... DDA-COCO consists of five subsets containing reconstructed images of MSCOCO [28] validation set by different VAEs, utilizing frequency-level alignment. |
| Hardware Specification | Yes | All experiments were conducted on eight NVIDIA V100 GPUs. |
| Software Dependencies | No | The paper states: 'We use DINOv2 as the backbone and fine-tune it with Lo RA, using a rank of 8.' and mentions 'VAE from Stable Diffusion 2.1'. While specific models are named, specific software dependencies like Python, PyTorch, or CUDA *with version numbers* are not provided. |
| Experiment Setup | Yes | We trained the detector on a dataset consisting of MSCOCO images and their synthetic counterparts generated through DDA alignment using the VAE from Stable Diffusion 2.1. The model was optimized with a base batch size of 16 and a learning rate of 1e-4. To achieve an effective batch size of 64 without exceeding GPU memory limits, gradient accumulation was applied over four iterations. Balanced accuracy was evaluated on all datasets every 10,000 iterations, and early stopping was employed to prevent overfitting. |