ForgDiffuser: General Image Forgery Localization with Diffusion Models

Authors: Mengxi Wang, Shaozhang Niu, Jiwei Zhang

IJCAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on six benchmark datasets demonstrate that Forg Diffuser outperforms existing mainstream GIFL methods in both localization accuracy and robustness, especially under challenging manipulation conditions. In this subsection, we conduct ablation experiments on the proposed Forg Diffuser to verify the effectiveness of each designed module. Quantitative comparisons: Table 1 demonstrates the quantitative results of Forg Diffuser with five baseline methods on six benchmark datasets.
Researcher Affiliation Academia 1Beijing Key Lab of Intelligent Telecommunication Software and Multimedia, Beijing University of Posts and Telecommunications, China 2Southeast Digital Economy Development Institute, China 3Key Laboratory of Interactive Technology and Experience System, Ministry of Culture and Tourism (BUPT), China EMAIL
Pseudocode No The paper describes the proposed method, Forg Diffuser, using textual explanations and diagrams (Figure 2 and Figure 3) rather than formal pseudocode or algorithm blocks.
Open Source Code No The paper does not contain any explicit statements or links indicating that the source code for the described methodology is publicly available.
Open Datasets Yes Forg Diffuser is evaluated on widely used forgery image datasets: CASIA1 [Dong et al., 2013], DID [Wu and Zhou, 2021], IMD [Novozamsky et al., 2020], Auto [Jia et al., 2023], BSN and RLS26K [Hao et al., 2024].
Dataset Splits Yes We divide the above datasets into train and test sets in the ratio of 9:1 for experimentation.
Hardware Specification Yes We implemented Forg Diffuser based on the Py Torch with one NVIDIA L20 with 48 GB memory for training and inference.
Software Dependencies No The paper mentions 'Py Torch' but does not specify its version or any other software dependencies with their version numbers.
Experiment Setup Yes We trained 100 epochs with batch sizes of 16. AGM is initialized using PVTv2-B4, and the input images are resized to 352 352. The Adam W optimizer is employed, and the initial learning rate is set to 0.001. λ1 and λ2 in the loss function of Equation 12 are set to 0.8 and 0.2, respectively. The time step T is set to 10 for sampling.