Search

The continuous advancement of face forgery techniques has caused a series of trust crises, posing a significant menace to information security and personal privacy. In response, deep learning is being employed to develop effective detection methods to identify deepfake images and videos. Currently, most detection methods generally achieve satisfactory performance in intra-domain detection. However, their effectiveness remains uncertain when detecting unknown forgeries and cross-domain scenarios. From a causal learning perspective, our analysis reveals that the failure of detectors in cross-domain detection primarily arises from spurious correlations between irrelevant features and category labels (real and fake). In this study, we propose the Feature Independence Constrainer (FIC) to alleviate spurious correlations and enhance the generalization for detection method. Specifically, FIC maps features to a Reproducing Kernel Hilbert Space and calculates the covariance matrix among the kernel-mapped features. Dynamic weight parameters are incorporated into the covariance matrix for iterative optimization. Consequently, statistical correlations between features are eliminated, ensuring independence between each pair of features and mitigating the impact of spurious correlations. Additionally, we introduce fine-grained high-frequency components to force the detection model to learn comprehensive forgery-related artifacts while avoiding spurious correlations between irrelevant features and labels. Furthermore, a Feature Alignment Module (FAM) is designed to explore higher-order dependencies between the spatial and frequency domains, allowing for the extraction of richer potential forgery traces. Quantitative and qualitative experiments demonstrate that the proposed method achieves competitive performance across multiple face forgery benchmark datasets, outperforming state-of-the-art methods., Competing Interests: Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper., (Copyright © 2024 Elsevier Ltd. All rights reserved.)

Artificial intelligence (AI) has been successful at solving numerous problems in machine perception. In radiology, AI systems are rapidly evolving and show progress in guiding treatment decisions, diagnosing, localizing disease on medical images, and improving radiologists' efficiency. A critical component to deploying AI in radiology is to gain confidence in a developed system's efficacy and safety. The current gold standard approach is to conduct an analytical validation of performance on a generalization dataset from one or more institutions, followed by a clinical validation study of the system's efficacy during deployment. Clinical validation studies are time-consuming, and best practices dictate limited re-use of analytical validation data, so it is ideal to know ahead of time if a system is likely to fail analytical or clinical validation. In this paper, we describe a series of sanity tests to identify when a system performs well on development data for the wrong reasons. We illustrate the sanity tests' value by designing a deep learning system to classify pancreatic cancer seen in computed tomography scans.