AIFIND: Artifact-Aware Interpreting Fine-Grained Alignment for Incremental Face Forgery Detection

TL;DR

AIFIND introduces artifact-aware semantic anchors and attention mechanisms to improve incremental face forgery detection, effectively reducing feature drift and catastrophic forgetting.

Contribution

It proposes a novel artifact-driven semantic prior generator and attention-based alignment method for more stable incremental learning in face forgery detection.

Findings

AIFIND outperforms existing methods on multiple incremental protocols.

Semantic anchors stabilize feature space during incremental learning.

The approach effectively mitigates catastrophic forgetting.

Abstract

As forgery types continue to emerge consistently, Incremental Face Forgery Detection (IFFD) has become a crucial paradigm. However, existing methods typically rely on data replay or coarse binary supervision, which fails to explicitly constrain the feature space, leading to severe feature drift and catastrophic forgetting. To address this, we propose AIFIND, Artifact-Aware Interpreting Fine-Grained Alignment for Incremental Face Forgery Detection, which leverages semantic anchors to stabilize incremental learning. We design the Artifact-Driven Semantic Prior Generator to instantiate invariant semantic anchors, establishing a fixed coordinate system from low-level artifact cues. These anchors are injected into the image encoder via Artifact-Probe Attention, which explicitly constrains volatile visual features to align with stable semantic anchors. Adaptive Decision Harmonizer harmonizes the classifiers by preserving angular relationships of semantic anchors, maintaining geometric consistency across tasks. Extensive experiments on multiple incremental protocols validate the superiority of AIFIND.