Learning Discriminative Noise Guidance for Image Forgery Detection and Localization

Authors: Jiaying Zhu, Dong Li, Xueyang Fu, Gang Yang, Jie Huang, Aiping Liu, Zheng-Jun Zha

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

Reproducibility Variable Result LLM Response
Research Type Experimental Comprehensive experiments on multiple datasets demonstrate that our method can reliably detect and localize forgeries, surpassing existing state-of-the-art methods.
Researcher Affiliation Academia University of Science and Technology of China {zhujy53, dongli6, yg1997}@mail.ustc.edu.cn, {xyfu, aipingl, zhazj}@ustc.edu.cn
Pseudocode Yes Algorithm 1: Cross-attention-based Guided Filter (CAGF) Input: Gn: guidance image(Noise), Gr: input image(RGB), Output: q: CAGF filtering output. 1: meann = MFConv(Gn) meanr = MFConv(Gr) 2: # Calculate variance and covariance (CMA: Figure 3) varn = CMA(Gn, Gn) covnr = CMA(Gn, Gr) 3: # Calculate coefficients of local linear relationship a = Res Block(Concat(covnr, varn)) b = meanr a meann 4: meana = MFConv(a) meanb = MFConv(b) 5: # Output q = meana * Gn + meanb return q
Open Source Code No The paper does not provide concrete access to source code for the methodology described.
Open Datasets Yes We evaluate our model on CASIA (Dong, Wang, and Tan 2013), Coverage (Wen et al. 2016), Columbia (Hsu and Chang 2006), Nist Nimble 2016 (NIST16) (Guan et al. 2019) and IMD20 (Novozamsky, Mahdian, and Saic 2020).
Dataset Splits No The paper mentions training and testing splits, but does not explicitly provide details about validation splits or the specific proportions of data used for training, validation, and testing.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory amounts, or detailed computer specifications) used for running its experiments.
Software Dependencies No The paper mentions various models and layers used (e.g., CBDNet, ResNet-50), but does not provide specific version numbers for these or any underlying software dependencies like programming languages, frameworks, or libraries.
Experiment Setup Yes Ln = λ (1 JSD) + (1 λ) L (Y, Gc) , (6) ... where λ is the hyperparameter to balance the two terms which is set as 0.80. ... LN = αL1 (Y, Gout)+βL2 (y, Dout)+(1 α β) L3 E , Ge , (13) ... In practice, α is set as 0.60 and β is set as 0.2.