A New Benchmark and Model for Challenging Image Manipulation Detection

Authors: Zhenfei Zhang, Mingyang Li, Ming-Ching Chang

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on the CIMD benchmark show that our model significantly outperforms So TA IMD methods on CIMD.
Researcher Affiliation Academia 1Department of Computer Science, University at Albany, State University of New York, New York, USA, 12222 2Department of Bioengineering, Mc Gill University, Montreal, QC, Canada, H3A 0E9
Pseudocode No The paper describes the method using text and mathematical equations, but does not include a formal pseudocode block or algorithm listing.
Open Source Code No The dataset is available at: https://github.com/Zhenfei Z/CIMD. The paper states the dataset is available at this link, but does not explicitly state that the source code for the methodology is also provided there or elsewhere.
Open Datasets Yes To investigate the State-of-The-Art (So TA) IMD methods in those challenging conditions, we introduce a new Challenging Image Manipulation Detection (CIMD) benchmark dataset, which consists of two subsets... The dataset is available at: https://github.com/Zhenfei Z/CIMD.
Dataset Splits No The training datasets used in this study were adopted from (Kwon et al. 2022). The testing phase entailed the utilization of CIMD-R and CIMD-C to evaluate the efficacy of image-editing-based and compression-based methods, respectively. However, specific training/validation/test splits for these datasets are not provided.
Hardware Specification Yes Our model was implemented using Py Torch (Paszke et al. 2019) and trained on 8 RTX 2080 GPUs, with batch size 4.
Software Dependencies Yes Our model was implemented using Py Torch (Paszke et al. 2019)
Experiment Setup Yes Our model was implemented using Py Torch (Paszke et al. 2019) and trained on 8 RTX 2080 GPUs, with batch size 4. We set the initial learning rate as 0.001 with exponential decay. The training process consists of 250 epochs.