LAKAN: Landmark-assisted Adaptive Kolmogorov-Arnold Network for Face Forgery Detection

TL;DR

This work introduces LAKAN, a landmark-assisted adaptive Kolmogorov-Arnold Network for face forgery detection. By replacing fixed activations with learnable edge-based splines and conditioning these activations on facial landmarks, LAKAN adaptively steers a general image encoder toward regions likely to contain forgery artifacts. The approach integrates a landmark-driven parameter generator with a dynamic KAN layer, producing instance-specific gating that highlights semantically important facial areas. Extensive cross-dataset and cross-manipulation experiments demonstrate strong generalization across diverse datasets and encoder backbones, validating the effectiveness of combining geometric priors with learned non-linear activations for forgery detection.

Abstract

The rapid development of deepfake generation techniques necessitates robust face forgery detection algorithms. While methods based on Convolutional Neural Networks (CNNs) and Transformers are effective, there is still room for improvement in modeling the highly complex and non-linear nature of forgery artifacts. To address this issue, we propose a novel detection method based on the Kolmogorov-Arnold Network (KAN). By replacing fixed activation functions with learnable splines, our KAN-based approach is better suited to this challenge. Furthermore, to guide the network's focus towards critical facial areas, we introduce a Landmark-assisted Adaptive Kolmogorov-Arnold Network (LAKAN) module. This module uses facial landmarks as a structural prior to dynamically generate the internal parameters of the KAN, creating an instance-specific signal that steers a general-purpose image encoder towards the most informative facial regions with artifacts. This core innovation creates a powerful combination between geometric priors and the network's learning process. Extensive experiments on multiple public datasets show that our proposed method achieves superior performance.

Figures (3)

• Figure 1: Compared to KAN DBLP:conf/iclr/LiuWVRHS0T25, LAKAN adaptively constructs activation functions based on facial landmark information, enabling it to make use of the key clues provided by landmark annotations fully.
• Figure 2: Overview of model architecture with LAKAN. The LAKAN module leverages facial landmarks to generate adaptive parameters for KAN and is applied to downsampled features from four different stages of the image encoder through gating mechanisms.
• Figure 3: Landmarks and saliency maps on four distinct manipulated samples and one real sample from the FF++ dataset.