InfoDense: Density-Aware Regional Decisive Replay for Memory-Efficient Incremental Face Forgery Detection

Authors: Jikang Cheng, Hao Shen, Xueyi Zhang, Guangcheng Wang, Zhongyuan Wang, Renye Yan, Baojin Huang

Published: 2026-07-18 16:25:27+00:00

AI Summary

InfoDense is a novel replay strategy for Incremental Face Forgery Detection (IFFD) that addresses catastrophic forgetting and memory limitations. It achieves this by prioritizing artifact-dense and forgery-critical regions of historical data, significantly reducing storage while preserving crucial forgery evidence. The method localizes decisive patches, ranks candidate segments for diversity and information density, and reconstructs unbiased training inputs by merging stored segments with current-task samples.

Abstract

The rapid evolution of face forgery techniques has introduced an increasing variety of manipulations. Incremental Face Forgery Detection (IFFD), which incrementally adds new forgery data to fine-tune previously trained models, has emerged as a promising approach to handle evolving forgery threats. However, conventional replay-based IFFD methods suffer from catastrophic forgetting. Storing full historical images under limited memory often either fails to preserve subtle forgery cues or introduces domain bias, reducing the model's ability to learn intrinsic and transferable manipulation characteristics. In this paper, we propose a Density-Aware Regional Decisive replay strategy, termed InfoDense, to address these challenges. InfoDense prioritizes artifact-dense and forgery-critical regions, significantly reducing storage requirements while maintaining high-fidelity forgery evidence. We first introduce InfoDense Cut to localize decisive patches using CLIP-based embeddings. Then, InfoDense Select ranks candidate segments by combining latent-space representativeness and decisive patch counts, ensuring both diversity and information density in the replay buffer. Finally, InfoDense Fuse reconstructs unbiased training inputs by adaptively merging stored segments with current-task samples, enhancing knowledge retention and generalization. Extensive experiments on challenging incremental deepfake benchmarks demonstrate that InfoDense effectively mitigates catastrophic forgetting while improving cross-domain generalization.

Key findings

InfoDense significantly mitigates catastrophic forgetting, as evidenced by the lowest Performance Dropping rate. It also substantially improves cross-domain generalization by focusing on intrinsic forgery cues. The method offers strong privacy preservation by disrupting facial structure in stored segments, making identity retrieval nearly impossible.

Approach

InfoDense employs a three-step process: First, InfoDense Cut localizes decisive patches using CLIP-based embeddings. Second, InfoDense Select ranks candidate segments based on latent-space representativeness and decisive patch counts to populate the replay buffer. Finally, InfoDense Fuse reconstructs training inputs by adaptively merging stored segments with current-task samples, preventing catastrophic forgetting and enhancing generalization.

Datasets

FaceForensics++ (FF++), DeepFake Detection Challenge Preview (DFDCP), Celeb-DF-v2 (CDF), DeepFake Detection (DFD), WildDeepfake (WDF), UADFV, DiT, SDv21 (from DiffusionFace and DF40).

Model(s)

CLIP ViT-L/14 (as backbone network) with a linear classifier and LoRA, ArcFace, CosFace.

Author countries

China, Singapore

← Previous