FakeI2V-Bench: Benchmarking the Applicability of Image-level Deepfake Detectors for Deepfake Video Detection

View on arXiv ← Back to list

Authors: Pei Li, Sihan Chen, Delong Ran, Tianshuo Cong

Published: 2026-08-04 04:13:31+00:00

Comment: To Appear in KDD 2026, Jeju, Korea, August 9-13, 2026

AI Summary

This paper introduces FakeI2V-Bench, a comprehensive benchmark for deepfake video detection that systematically evaluates both video-level and image-level detectors. The authors also propose IV-Bridge, a framework to enhance the applicability of image-level deepfake detectors for video tasks. Their findings indicate that enhanced image-level detectors can surpass state-of-the-art video-level methods, even with lower computational costs.

Abstract

Recent advances in video generation models have significantly intensified the deepfake threat, yet the current deepfake video detection benchmarks remain underdeveloped. In particular, the effectiveness of image-level detectors in the video domain has not been systematically assessed. To fill this gap, we present FakeI2V-Bench, a benchmark for evaluating state-of-the-art video-level deepfake detectors in challenging scenarios, with a particular focus on systematically assessing the performance of image-level deepfake detectors in the video domain. FakeI2V-Bench comprises 97,548 videos, containing content generated by the latest powerful generation models and covering a broader range of categories. Using this dataset, we conduct a systematic evaluation of eight video-level detectors and twelve representative image-level detectors. Experimental results show that the best-performing image-level detector achieves an 80.16% AUC, slightly outperforming the strongest video-level detector (i.e., 79.99% AUC). Going beyond benchmarking, we present IV-Bridge, a general framework that enhances the applicability of image-level deepfake detectors to videos. IV-Bridge employs a random forest model with statistical features to aggregate frame-level predictions, allowing eleven image-level detectors to surpass state-of-the-art video-level approaches, with the best-performing variant achieving a 93.80% AUC. Overall, FakeI2V-Bench establishes a rigorous benchmark for deepfake video detection and introduces a novel pathway for extending image-level detectors to the video domain, offering new insights and directions for future research. Code and data are available at https://github.com/CryptoAILab/FakeI2V-Bench.

Key findings
The best-performing naive image-level detector achieved an 80.16% AUC, slightly outperforming the strongest video-level detector (79.99% AUC) after optimal frame-to-video aggregation. With the IV-Bridge framework, eleven out of twelve enhanced image-level detectors surpassed state-of-the-art video-level approaches, with the best variant achieving a 93.80% AUC, demonstrating stronger generalization and lower computational costs.
Approach
The study first establishes FakeI2V-Bench, a dataset of 97,548 videos, to evaluate video-level and naive image-level deepfake detectors. It then introduces IV-Bridge, a two-stage framework involving Video-Frame Fine-Tuning (VFT) to adapt image-level detectors to video-specific patterns and Multi-Mode Aggregation (MMA) using a Random Forest model to combine frame-level predictions into a video-level decision.
Datasets
FakeI2V-Bench (comprising Celeb-DF v2, DeepFake Detection (DFD), GenVideo (GV), GenVidBench (GVB)), FakeGenImage (for image-level detector evaluation), FF++ (for VFT), MSR-VTT (for real videos in GV), Vript (for real videos in GVB).
Model(s)
Video-level detectors: LAA-Net, FTCN, UIA-ViT, AltFreezing, MM-Det, VGMShield, DeMamba, D3. Image-level detectors: CNNDet, Patch, LNP, LGrad, DeFake, DIRE, DMID, CoDE, DRCT, UniFD, NPR, RINE. Enhancement framework: Random Forest (for Multi-Mode Aggregation).
Author countries
China
← Previous
Next →