The rise of deepfake images, especially of well-known personalities, poses a serious threat to the dissemination of authentic information. To tackle this, we present a thorough investigation into how deepfakes are produced and how they can be identified. The cornerstone of our research is a rich collection of artificial celebrity faces, titled DeepFakeFace (DFF). We crafted the DFF dataset using advanced diffusion models and have shared it with the community through online platforms. This data serves as a robust foundation to train and test algorithms designed to spot deepfakes. We carried out a thorough review of the DFF dataset and suggest two evaluation methods to gauge the strength and adaptability of deepfake recognition tools. The first method tests whether an algorithm trained on one type of fake images can recognize those produced by other methods. The second evaluates the algorithm's performance with imperfect images, like those that are blurry, of low quality, or compressed. Given varied results across deepfake methods and image changes, our findings stress the need for better deepfake detectors. Our DFF dataset and tests aim to boost the development of more effective tools against deepfakes.
Deepfakes are becoming increasingly credible, posing a significant threat given their potential to facilitate fraud or bypass access control systems. This has motivated the development of deepfake detection methods, in which deep learning models are trained to distinguish between real and synthesized footage. Unfortunately, existing detectors struggle to generalize to deepfakes from datasets they were not trained on, but little work has been done to examine why or how this limitation can be addressed. Especially, those single-modality deepfake images reveal little available forgery evidence, posing greater challenges than detecting deepfake videos. In this work, we present the first empirical study on the generalizability of deepfake detectors, an essential goal for detectors to stay one step ahead of attackers. Our study utilizes six deepfake datasets, five deepfake image detection methods, and two model augmentation approaches, confirming that detectors do not generalize in zero-shot settings. Additionally, we find that detectors are learning unwanted properties specific to synthesis methods and struggling to extract discriminative features, limiting their ability to generalize. Finally, we find that there are neurons universally contributing to detection across seen and unseen datasets, suggesting a possible path towards zero-shot generalizability.
Large Language Models (LLMs) have achieved human-level fluency in text generation, making it difficult to distinguish between human-written and LLM-generated texts. This poses a growing risk of misuse of LLMs and demands the development of detectors to identify LLM-generated texts. However, existing detectors lack robustness against attacks: they degrade detection accuracy by simply paraphrasing LLM-generated texts. Furthermore, a malicious user might attempt to deliberately evade the detectors based on detection results, but this has not been assumed in previous studies. In this paper, we propose OUTFOX, a framework that improves the robustness of LLM-generated-text detectors by allowing both the detector and the attacker to consider each other's output. In this framework, the attacker uses the detector's prediction labels as examples for in-context learning and adversarially generates essays that are harder to detect, while the detector uses the adversarially generated essays as examples for in-context learning to learn to detect essays from a strong attacker. Experiments in the domain of student essays show that the proposed detector improves the detection performance on the attacker-generated texts by up to +41.3 points F1-score. Furthermore, the proposed detector shows a state-of-the-art detection performance: up to 96.9 points F1-score, beating existing detectors on non-attacked texts. Finally, the proposed attacker drastically degrades the performance of detectors by up to -57.0 points F1-score, massively outperforming the baseline paraphrasing method for evading detection.
Deepfakes have raised significant concerns due to their potential to spread false information and compromise digital media integrity. In this work, we propose a Generative Convolutional Vision Transformer (GenConViT) for deepfake video detection. Our model combines ConvNeXt and Swin Transformer models for feature extraction, and it utilizes Autoencoder and Variational Autoencoder to learn from the latent data distribution. By learning from the visual artifacts and latent data distribution, GenConViT achieves improved performance in detecting a wide range of deepfake videos. The model is trained and evaluated on DFDC, FF++, DeepfakeTIMIT, and Celeb-DF v2 datasets, achieving high classification accuracy, F1 scores, and AUC values. The proposed GenConViT model demonstrates robust performance in deepfake video detection, with an average accuracy of 95.8% and an AUC value of 99.3% across the tested datasets. Our proposed model addresses the challenge of generalizability in deepfake detection by leveraging visual and latent features and providing an effective solution for identifying a wide range of fake videos while preserving media integrity. The code for GenConViT is available at https://github.com/erprogs/GenConViT.
Despite the development of effective deepfake detectors in recent years, recent studies have demonstrated that biases in the data used to train these detectors can lead to disparities in detection accuracy across different races and genders. This can result in different groups being unfairly targeted or excluded from detection, allowing undetected deepfakes to manipulate public opinion and erode trust in a deepfake detection model. While existing studies have focused on evaluating fairness of deepfake detectors, to the best of our knowledge, no method has been developed to encourage fairness in deepfake detection at the algorithm level. In this work, we make the first attempt to improve deepfake detection fairness by proposing novel loss functions that handle both the setting where demographic information (eg, annotations of race and gender) is available as well as the case where this information is absent. Fundamentally, both approaches can be used to convert many existing deepfake detectors into ones that encourages fairness. Extensive experiments on four deepfake datasets and five deepfake detectors demonstrate the effectiveness and flexibility of our approach in improving deepfake detection fairness. Our code is available at https://github.com/littlejuyan/DF_Fairness.
The rapid advancements in computer vision have stimulated remarkable progress in face forgery techniques, capturing the dedicated attention of researchers committed to detecting forgeries and precisely localizing manipulated areas. Nonetheless, with limited fine-grained pixel-wise supervision labels, deepfake detection models perform unsatisfactorily on precise forgery detection and localization. To address this challenge, we introduce the well-trained vision segmentation foundation model, i.e., Segment Anything Model (SAM) in face forgery detection and localization. Based on SAM, we propose the Detect Any Deepfakes (DADF) framework with the Multiscale Adapter, which can capture short- and long-range forgery contexts for efficient fine-tuning. Moreover, to better identify forged traces and augment the model's sensitivity towards forgery regions, Reconstruction Guided Attention (RGA) module is proposed. The proposed framework seamlessly integrates end-to-end forgery localization and detection optimization. Extensive experiments on three benchmark datasets demonstrate the superiority of our approach for both forgery detection and localization. The codes will be released soon at https://github.com/laiyingxin2/DADF.
With a recent influx of voice generation methods, the threat introduced by audio DeepFake (DF) is ever-increasing. Several different detection methods have been presented as a countermeasure. Many methods are based on so-called front-ends, which, by transforming the raw audio, emphasize features crucial for assessing the genuineness of the audio sample. Our contribution contains investigating the influence of the state-of-the-art Whisper automatic speech recognition model as a DF detection front-end. We compare various combinations of Whisper and well-established front-ends by training 3 detection models (LCNN, SpecRNet, and MesoNet) on a widely used ASVspoof 2021 DF dataset and later evaluating them on the DF In-The-Wild dataset. We show that using Whisper-based features improves the detection for each model and outperforms recent results on the In-The-Wild dataset by reducing Equal Error Rate by 21%.
Existing deepfake detection methods fail to generalize well to unseen or degraded samples, which can be attributed to the over-fitting of low-level forgery patterns. Here we argue that high-level semantics are also indispensable recipes for generalizable forgery detection. Recently, large pre-trained Vision Transformers (ViTs) have shown promising generalization capability. In this paper, we propose the first parameter-efficient tuning approach for deepfake detection, namely DeepFake-Adapter, to effectively and efficiently adapt the generalizable high-level semantics from large pre-trained ViTs to aid deepfake detection. Given large pre-trained models but limited deepfake data, DeepFake-Adapter introduces lightweight yet dedicated dual-level adapter modules to a ViT while keeping the model backbone frozen. Specifically, to guide the adaptation process to be aware of both global and local forgery cues of deepfake data, 1) we not only insert Globally-aware Bottleneck Adapters in parallel to MLP layers of ViT, 2) but also actively cross-attend Locally-aware Spatial Adapters with features from ViT. Unlike existing deepfake detection methods merely focusing on low-level forgery patterns, the forgery detection process of our model can be regularized by generalizable high-level semantics from a pre-trained ViT and adapted by global and local low-level forgeries of deepfake data. Extensive experiments on several standard deepfake detection benchmarks validate the effectiveness of our approach. Notably, DeepFake-Adapter demonstrates a convincing advantage under cross-dataset and cross-manipulation settings. The code has been released at https://github.com/rshaojimmy/DeepFake-Adapter.
Audio Deepfake Detection (ADD) aims to detect the fake audio generated by text-to-speech (TTS), voice conversion (VC) and replay, etc., which is an emerging topic. Traditionally we take the mono signal as input and focus on robust feature extraction and effective classifier design. However, the dual-channel stereo information in the audio signal also includes important cues for deepfake, which has not been studied in the prior work. In this paper, we propose a novel ADD model, termed as M2S-ADD, that attempts to discover audio authenticity cues during the mono-to-stereo conversion process. We first projects the mono to a stereo signal using a pretrained stereo synthesizer, then employs a dual-branch neural architecture to process the left and right channel signals, respectively. In this way, we effectively reveal the artifacts in the fake audio, thus improve the ADD performance. The experiments on the ASVspoof2019 database show that M2S-ADD outperforms all baselines that input mono. We release the source code at \url{https://github.com/AI-S2-Lab/M2S-ADD}.
The recent renaissance in generative models, driven primarily by the advent of diffusion models and iterative improvement in GAN methods, has enabled many creative applications. However, each advancement is also accompanied by a rise in the potential for misuse. In the arena of the deepfake generation, this is a key societal issue. In particular, the ability to modify segments of videos using such generative techniques creates a new paradigm of deepfakes which are mostly real videos altered slightly to distort the truth. This paradigm has been under-explored by the current deepfake detection methods in the academic literature. In this paper, we present a deepfake detection method that can address this issue by performing deepfake prediction at the frame and video levels. To facilitate testing our method, we prepared a new benchmark dataset where videos have both real and fake frame sequences with very subtle transitions. We provide a benchmark on the proposed dataset with our detection method which utilizes the Vision Transformer based on Scaling and Shifting to learn spatial features, and a Timeseries Transformer to learn temporal features of the videos to help facilitate the interpretation of possible deepfakes. Extensive experiments on a variety of deepfake generation methods show excellent results by the proposed method on temporal segmentation and classical video-level predictions as well. In particular, the paradigm we address will form a powerful tool for the moderation of deepfakes, where human oversight can be better targeted to the parts of videos suspected of being deepfakes. All experiments can be reproduced at: github.com/rgb91/temporal-deepfake-segmentation.
Malicious Deepfakes have led to a sharp conflict over distinguishing between genuine and forged faces. Although many countermeasures have been developed to detect Deepfakes ex-post, undoubtedly, passive forensics has not considered any preventive measures for the pristine face before foreseeable manipulations. To complete this forensics ecosystem, we thus put forward the proactive solution dubbed SepMark, which provides a unified framework for source tracing and Deepfake detection. SepMark originates from encoder-decoder-based deep watermarking but with two separable decoders. For the first time the deep separable watermarking, SepMark brings a new paradigm to the established study of deep watermarking, where a single encoder embeds one watermark elegantly, while two decoders can extract the watermark separately at different levels of robustness. The robust decoder termed Tracer that resists various distortions may have an overly high level of robustness, allowing the watermark to survive both before and after Deepfake. The semi-robust one termed Detector is selectively sensitive to malicious distortions, making the watermark disappear after Deepfake. Only SepMark comprising of Tracer and Detector can reliably trace the trusted source of the marked face and detect whether it has been altered since being marked; neither of the two alone can achieve this. Extensive experiments demonstrate the effectiveness of the proposed SepMark on typical Deepfakes, including face swapping, expression reenactment, and attribute editing.
Most deepfake detection methods focus on detecting spatial and/or spatio-temporal changes in facial attributes and are centered around the binary classification task of detecting whether a video is real or fake. This is because available benchmark datasets contain mostly visual-only modifications present in the entirety of the video. However, a sophisticated deepfake may include small segments of audio or audio-visual manipulations that can completely change the meaning of the video content. To addresses this gap, we propose and benchmark a new dataset, Localized Audio Visual DeepFake (LAV-DF), consisting of strategic content-driven audio, visual and audio-visual manipulations. The proposed baseline method, Boundary Aware Temporal Forgery Detection (BA-TFD), is a 3D Convolutional Neural Network-based architecture which effectively captures multimodal manipulations. We further improve (i.e. BA-TFD+) the baseline method by replacing the backbone with a Multiscale Vision Transformer and guide the training process with contrastive, frame classification, boundary matching and multimodal boundary matching loss functions. The quantitative analysis demonstrates the superiority of BA-TFD+ on temporal forgery localization and deepfake detection tasks using several benchmark datasets including our newly proposed dataset. The dataset, models and code are available at https://github.com/ControlNet/LAV-DF.
Deepfakes refer to content synthesized using deep generators, which, when misused, have the potential to erode trust in digital media. Synthesizing high-quality deepfakes requires access to large and complex generators only a few entities can train and provide. The threat is malicious users that exploit access to the provided model and generate harmful deepfakes without risking detection. Watermarking makes deepfakes detectable by embedding an identifiable code into the generator that is later extractable from its generated images. We propose Pivotal Tuning Watermarking (PTW), a method for watermarking pre-trained generators (i) three orders of magnitude faster than watermarking from scratch and (ii) without the need for any training data. We improve existing watermarking methods and scale to generators $4 \times$ larger than related work. PTW can embed longer codes than existing methods while better preserving the generator's image quality. We propose rigorous, game-based definitions for robustness and undetectability, and our study reveals that watermarking is not robust against an adaptive white-box attacker who controls the generator's parameters. We propose an adaptive attack that can successfully remove any watermarking with access to only 200 non-watermarked images. Our work challenges the trustworthiness of watermarking for deepfake detection when the parameters of a generator are available. The source code to reproduce our experiments is available at https://github.com/nilslukas/gan-watermark.
Misinformation has become a pressing issue. Fake media, in both visual and textual forms, is widespread on the web. While various deepfake detection and text fake news detection methods have been proposed, they are only designed for single-modality forgery based on binary classification, let alone analyzing and reasoning subtle forgery traces across different modalities. In this paper, we highlight a new research problem for multi-modal fake media, namely Detecting and Grounding Multi-Modal Media Manipulation (DGM^4). DGM^4 aims to not only detect the authenticity of multi-modal media, but also ground the manipulated content (i.e., image bounding boxes and text tokens), which requires deeper reasoning of multi-modal media manipulation. To support a large-scale investigation, we construct the first DGM^4 dataset, where image-text pairs are manipulated by various approaches, with rich annotation of diverse manipulations. Moreover, we propose a novel HierArchical Multi-modal Manipulation rEasoning tRansformer (HAMMER) to fully capture the fine-grained interaction between different modalities. HAMMER performs 1) manipulation-aware contrastive learning between two uni-modal encoders as shallow manipulation reasoning, and 2) modality-aware cross-attention by multi-modal aggregator as deep manipulation reasoning. Dedicated manipulation detection and grounding heads are integrated from shallow to deep levels based on the interacted multi-modal information. Finally, we build an extensive benchmark and set up rigorous evaluation metrics for this new research problem. Comprehensive experiments demonstrate the superiority of our model; several valuable observations are also revealed to facilitate future research in multi-modal media manipulation.
Recent advancements in diffusion models have enabled the generation of realistic deepfakes from textual prompts in natural language. While these models have numerous benefits across various sectors, they have also raised concerns about the potential misuse of fake images and cast new pressures on fake image detection. In this work, we pioneer a systematic study on deepfake detection generated by state-of-the-art diffusion models. Firstly, we conduct a comprehensive analysis of the performance of contrastive and classification-based visual features, respectively extracted from CLIP-based models and ResNet or ViT-based architectures trained on image classification datasets. Our results demonstrate that fake images share common low-level cues, which render them easily recognizable. Further, we devise a multimodal setting wherein fake images are synthesized by different textual captions, which are used as seeds for a generator. Under this setting, we quantify the performance of fake detection strategies and introduce a contrastive-based disentangling method that lets us analyze the role of the semantics of textual descriptions and low-level perceptual cues. Finally, we release a new dataset, called COCOFake, containing about 1.2M images generated from the original COCO image-caption pairs using two recent text-to-image diffusion models, namely Stable Diffusion v1.4 and v2.0.
A fast-paced development of DeepFake generation techniques challenge the detection schemes designed for known type DeepFakes. A reliable Deepfake detection approach must be agnostic to generation types, which can present diverse quality and appearance. Limited generalizability across different generation schemes will restrict the wide-scale deployment of detectors if they fail to handle unseen attacks in an open set scenario. We propose a new approach, Multi-Channel Xception Attention Pairwise Interaction (MCX-API), that exploits the power of pairwise learning and complementary information from different color space representations in a fine-grained manner. We first validate our idea on a publicly available dataset in a intra-class setting (closed set) with four different Deepfake schemes. Further, we report all the results using balanced-open-set-classification (BOSC) accuracy in an inter-class setting (open-set) using three public datasets. Our experiments indicate that our proposed method can generalize better than the state-of-the-art Deepfakes detectors. We obtain 98.48% BOSC accuracy on the FF++ dataset and 90.87% BOSC accuracy on the CelebDF dataset suggesting a promising direction for generalization of DeepFake detection. We further utilize t-SNE and attention maps to interpret and visualize the decision-making process of our proposed network. https://github.com/xuyingzhongguo/MCX-API
Deep-learning based face-swap videos, also known as deep fakes, are becoming more and more realistic and deceiving. The malicious usage of these face-swap videos has caused wide concerns. The research community has been focusing on the automatic detection of these fake videos, but the assessment of their visual realism, as perceived by human eyes, is still an unexplored dimension. Visual realism assessment, or VRA, is essential for assessing the potential impact that may be brought by a specific face-swap video, and it is also important as a quality assessment metric to compare different face-swap methods. In this paper, we make a small step towards this new VRA direction by building a benchmark for evaluating the effectiveness of different automatic VRA models, which range from using traditional hand-crafted features to different kinds of deep-learning features. The evaluations are based on a recent competition dataset named DFGC 2022, which contains 1400 diverse face-swap videos that are annotated with Mean Opinion Scores (MOS) on visual realism. Comprehensive experiment results using 11 models and 3 protocols are shown and discussed. We demonstrate the feasibility of devising effective VRA models for assessing face-swap videos and methods. The particular usefulness of existing deepfake detection features for VRA is also noted. The code can be found at https://github.com/XianyunSun/VRA.git.
Manipulated videos often contain subtle inconsistencies between their visual and audio signals. We propose a video forensics method, based on anomaly detection, that can identify these inconsistencies, and that can be trained solely using real, unlabeled data. We train an autoregressive model to generate sequences of audio-visual features, using feature sets that capture the temporal synchronization between video frames and sound. At test time, we then flag videos that the model assigns low probability. Despite being trained entirely on real videos, our model obtains strong performance on the task of detecting manipulated speech videos. Project site: https://cfeng16.github.io/audio-visual-forensics
Audio DeepFakes (DF) are artificially generated utterances created using deep learning, with the primary aim of fooling the listeners in a highly convincing manner. Their quality is sufficient to pose a severe threat in terms of security and privacy, including the reliability of news or defamation. Multiple neural network-based methods to detect generated speech have been proposed to prevent the threats. In this work, we cover the topic of adversarial attacks, which decrease the performance of detectors by adding superficial (difficult to spot by a human) changes to input data. Our contribution contains evaluating the robustness of 3 detection architectures against adversarial attacks in two scenarios (white-box and using transferability) and enhancing it later by using adversarial training performed by our novel adaptive training. Moreover, one of the investigated architectures is RawNet3, which, to the best of our knowledge, we adapted for the first time to DeepFake detection.
Deepfakes are computationally-created entities that falsely represent reality. They can take image, video, and audio modalities, and pose a threat to many areas of systems and societies, comprising a topic of interest to various aspects of cybersecurity and cybersafety. In 2020 a workshop consulting AI experts from academia, policing, government, the private sector, and state security agencies ranked deepfakes as the most serious AI threat. These experts noted that since fake material can propagate through many uncontrolled routes, changes in citizen behaviour may be the only effective defence. This study aims to assess human ability to identify image deepfakes of human faces (StyleGAN2:FFHQ) from nondeepfake images (FFHQ), and to assess the effectiveness of simple interventions intended to improve detection accuracy. Using an online survey, 280 participants were randomly allocated to one of four groups: a control group, and 3 assistance interventions. Each participant was shown a sequence of 20 images randomly selected from a pool of 50 deepfake and 50 real images of human faces. Participants were asked if each image was AI-generated or not, to report their confidence, and to describe the reasoning behind each response. Overall detection accuracy was only just above chance and none of the interventions significantly improved this. Participants' confidence in their answers was high and unrelated to accuracy. Assessing the results on a per-image basis reveals participants consistently found certain images harder to label correctly, but reported similarly high confidence regardless of the image. Thus, although participant accuracy was 62% overall, this accuracy across images ranged quite evenly between 85% and 30%, with an accuracy of below 50% for one in every five images. We interpret the findings as suggesting that there is a need for an urgent call to action to address this threat.