This is a page that curates AI-related papers published worldwide. All content here is summarized using Google Gemini and operated on a non-profit basis. Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.
This paper introduces the PLADA framework, which is proposed to address the issues that arise as images (deepfakes) generated by generative adversarial networks (GANs) and diffusion models (DMs) become difficult to distinguish from real images and are widely shared on online social networks (OSNs). PLADA overcomes the limitations of existing methods by considering the blocking effect due to compression of OSNs and effectively processes unpaired data and compressed images. The core module, Block Effect Remover (B2E), handles the blocking effect using a dual-stage attention mechanism, and Open Data Aggregation (ODA) improves the detection performance by processing both paired and unpaired data. Experimental results on 26 datasets show that PLADA outperforms state-of-the-art (SoTA) methods in deepfakes detection of OSNs even in limited paired data and compressed environments. This paper suggests blocking effect as an important factor in deepfakes detection and provides a robust solution for open-world scenarios.
Takeaways, Limitations
•
Takeaways:
◦
We present the blocking effect due to compression of OSNs as an important factor in deepfake detection.
◦
A PLADA framework is proposed to effectively handle unpaired data and compressed images.
◦
Achieving excellent deepfake detection performance even with limited paired data.
◦
Improving deepfake detection performance in OSNs environments.
◦
Providing a robust deepfake detection solution for open world scenarios.
◦
Reproducibility and extensibility through open code.
•
Limitations:
◦
Generalized performance evaluation for various compression methods and compression ratios is needed.
◦
Adaptability evaluation of new deepfake generation techniques is needed.
◦
Large-scale experiments and validation in real OSNs environments are needed.
◦
Possible lack of consideration of factors other than block effects (e.g. noise, distortion).
