DF-CBM: Region-Aware Concept Bottleneck Models for Deepfake Detection
DF-CBM is a region-aware concept bottleneck model that explains deepfake decisions with spatially grounded manipulation concepts.
DF-CBM builds a compact vocabulary of manipulation concepts from textual artifact annotations and ties each concept to facial and boundary regions. Concept-specific masked attention, guided by parsed facial masks, predicts those concepts from visual features, and the real-or-fake decision is made from the concept bottleneck. The model beats concept-based baselines on concept prediction and deepfake classification while remaining competitive with black-box detectors. Intervention analysis supports counterfactual explanations of how individual concepts change the prediction, and code is released on GitHub.
- Links manipulation concepts to facial and boundary regions.
- Masked attention uses parsed facial masks for concept prediction.
- Outperforms concept baselines and matches black-box detectors.
- Supports spatially grounded and counterfactual explanations.
From the original article
1 min read at arxiv.orgDeepfake detection methods have become increasingly effective yet most provide limited insight into the evidence behind their predictions. However, in forensic settings users also need to know which manipulation cues support the decision and where they appear.
By Georgios Tsoumplekas · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.