CCDF: A Benchmark Dataset for Deepfake Detection in Real-World Surveillance Footage

Authors: Baptiste Chopin, Thomas Swearingen, Arun Ross, Antitza Dantcheva, Christian Rathgeb

Published: 2026-10-06 08:13:10+00:00

AI Summary

This paper introduces CCDF, a new deepfake video dataset specifically designed for detecting synthetic content in real-world surveillance footage. It addresses limitations of existing datasets by focusing on plausible malicious content (crime and accident scenarios) generated by state-of-the-art commercial systems like Grok Imagine, Google VEO 3.1, and OpenAI Sora 2. Evaluation with ten recent deepfake detectors reveals that current methods struggle to reliably distinguish generated videos in CCDF from real ones, highlighting a significant gap in current detection capabilities for advanced generative AI.

Abstract

Due to rapid advances in Generative AI, commercial video generation tools can be used to produce fabricated surveillance footage that can fool both human viewers and automated synthetic video detectors. Since these tools are so widely accessible, a malicious user can create a harmful video clip at minimal cost. The production and dissemination of such videos in high-stakes settings, such as crime reporting and elections, can misdirect emergency response efforts or distort political discourse. Existing deepfake video datasets, used by the research community to develop deepfake detection algorithms, exhibit two limitations: (1) they emphasize benign web content rather than footage of possibly malicious activity, and (2) they rely on older or open-source generators that do not represent recent advances in generative systems. We assemble CCtv DeepFakes (CCDF), a video deepfake dataset, to address both gaps. CCDF contains 1840 videos (460 real and 1380 generated) spanning 16 crime and accident categories, with generated content produced using three leading commercial systems: Grok Imagine, Google VEO 3.1, and OpenAI Sora 2. CCDF is a highly realistic, small-scale, manually annotated dataset targeting evaluation of detection models. We release three versions of the dataset: the raw generated data, a cleaned version in which video metadata are standardized between real and synthetic samples to prevent detectors from exploiting trivial cues, and an altered version simulating low-effort post-processing attacks. We evaluate CCDF with ten recent state-of-the-art detectors covering different detection approaches. Our results suggest that these approaches do not reliably distinguish CCDF's generated videos from real ones, despite their strong reported performance on existing datasets. These results further confirm that existing datasets are not well-suited to evaluating certain threats.

Key findings

Current state-of-the-art deepfake detectors, despite strong performance on existing benchmarks, do not reliably distinguish between real and generated videos in the CCDF dataset, especially the 'cleaned' version. This indicates that existing datasets are not well-suited to evaluate threats from advanced commercial generative systems and that detectors rely on biases in older datasets which are removed in CCDF. The study highlights the urgent need for more robust detection methods and datasets that reflect high-stakes operational scenarios and frontier generative models.

Approach

The authors construct the CCDF dataset by using text prompts derived from real surveillance footage of crimes and accidents, which are then used to generate synthetic videos with leading commercial generative AI models. To prevent detectors from exploiting trivial cues, three versions of the dataset are released: raw, a 'cleaned' version with standardized video metadata, and an 'altered' version simulating post-processing attacks. Ten state-of-the-art deepfake detectors are then benchmarked against this dataset in a zero-shot setting.

Datasets

CCtv DeepFakes (CCDF), UCF-Crime, XD-Violence, MSAD, GenVideo

Model(s)

Grok Imagine, Google VEO 3.1, OpenAI Sora 2 (for generation); C2P-CLIP, ForgeLens, D3-Im, B-Free, Community Forensics, WaveRep, L3DE, NSG-VD, D3-Vid, STALL (for detection)

Author countries

Germany,United States,France

← Previous