Reproduce “SPARK-IL: Spectral Retrieval-Augmented RAG for Knowledge-driven Deepfake Detection via Incremental Learning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Hessen Bougueffa Eutamene, Abdellah Zakaria Sellam, Abdelmalik Taleb-Ahmed, Abdenour Hadid
The SPARK-IL framework detects deepfakes by leveraging stable frequency-domain characteristics through a dual-path multi-band spectral analysis, integrating retrieval-augmented inference and an incremental learning strategy. It achieved 94.60% mean accuracy on the UniversalFakeDetect benchmark, surpassing prior state-of-the-art methods by 0.70%.
Key insights
- Frequency-domain statistics exhibit more consistent deviations between real and synthetic images across different generative models than pixel-level artifacts.
- A dual-path approach combining pixel-level and semantic-level spectral analysis, processed by band-specific KANs, captures a comprehensive range of deepfake signatures.
- Integrating retrieval-augmented inference and incremental learning enables a robust and adaptive deepfake detection system that generalizes to unseen generators and mitigates catastrophic forgetting.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.