WWW: Where, Which and Whatever Enhancing Interpretability in Multimodal Deepfake Detection

TL;DR

This paper introduces FakeMix, a new benchmark and evaluation metrics for detecting localized deepfake manipulations in videos and audio, revealing limitations of current models in real-world scenarios.

Contribution

The paper presents FakeMix, a clip-level deepfake benchmark, and novel metrics TA and FDM, to improve interpretability and robustness assessment of detection models.

Findings

State-of-the-art models perform poorly on FakeMix with accuracy dropping to ~52%.

Existing benchmarks overestimate detection performance at the video level.

Proposed metrics effectively highlight model weaknesses in real-world deepfake detection.

Abstract

All current benchmarks for multimodal deepfake detection manipulate entire frames using various generation techniques, resulting in oversaturated detection accuracies exceeding 94% at the video-level classification. However, these benchmarks struggle to detect dynamic deepfake attacks with challenging frame-by-frame alterations presented in real-world scenarios. To address this limitation, we introduce FakeMix, a novel clip-level evaluation benchmark aimed at identifying manipulated segments within both video and audio, providing insight into the origins of deepfakes. Furthermore, we propose novel evaluation metrics, Temporal Accuracy (TA) and Frame-wise Discrimination Metric (FDM), to assess the robustness of deepfake detection models. Evaluating state-of-the-art models against diverse deepfake benchmarks, particularly FakeMix, demonstrates the effectiveness of our approach comprehensively. Specifically, while achieving an Average Precision (AP) of 94.2% at the video-level, the evaluation of the existing models at the clip-level using the proposed metrics, TA and FDM, yielded sharp declines in accuracy to 53.1%, and 52.1%, respectively.