CUPID: Reconstructing UV Texture Maps for Interpretable Person-of-Interest Deepfake Detection

2026-06-18

AI Summary

Existing deepfake detection methods struggle to balance robustness, efficiency, and interpretability, particularly when targeting high-profile persons of interest (POIs) under conditions of scarce or absent forged samples and identity-specific training data. This work proposes an unsupervised approach that requires no forged data, leveraging UV texture maps derived from 3D face reconstruction and integrating them with a Masked Autoencoder (MAE) for context-guided reconstruction. The method constructs a discriminative feature space generalizable to unseen identities and determines authenticity by measuring embedding discrepancies between query videos and original reference videos. By introducing UV textures and MAE into POI-focused forgery detection for the first time, the approach outperforms state-of-the-art methods across four benchmark datasets, demonstrates superior robustness against downsampling and compression attacks, achieves faster inference, and offers facial region-level interpretability through decoder residual maps.

Abstract

Deepfakes targeting a high-profile individual, known as Person-of-Interest (POI), are a threat to modern democracies and societies. Current POI deepfake detection methods still struggle to combine robustness to post-processing, efficiency and interpretability, focal aspects of modern deepfake detectors. In this paper we propose CUPID, a POI video deepfake detector that combines UV texture maps, a facial appearance representation derived from 3D face reconstructions, with the representation learning capabilities of the Masked Autoencoder (MAE). Our method does not require any deepfake videos in its training phase. Moreover, it does not even require to include a specific POI in the training set: the combination of UV texture maps extracted from real video frames and the MAE context-guided reconstruction yields a latent space that captures rich and discriminative facial features also for identities unseen during training. In the testing phase, the embeddings extracted from a query video depicting the POI can be matched against pristine reference videos to assess the video authenticity. Furthermore, operating in the UV space naturally provides an additional layer of interpretability. Specifically, we can extract decoded residual maps that highlight which facial regions of a test video deviate most from the identity representation of the corresponding POI. Experiments on four deepfake datasets show that CUPID outperforms current state of the art on most datasets and achieves the best overall robustness against strong downscaling and compression, providing also substantially faster inference. Our experimental code will be released at https://github.com/polimi-ispl/CUPID.

Problem

• Deepfake Detection
• Person-of-Interest
• Interpretability
• Robustness
• UV Texture Maps

Innovation

• UV texture maps
• Masked Autoencoder
• Person-of-Interest deepfake detection
• zero-shot identity generalization
• interpretable residual maps