Does data augmentation help or hinder the generalization of deepfake video detection?

With advances in generation techniques, malicious users can easily generate deepfake videos, which can cause severe social problems and trust issues. Therefore, deepfake video detection has received increasing attention in recent years. Given that forgery clues are often subtle and imperceptible, effective detection relies heavily on multi-grained learning. However, existing approaches fail to systematically incorporate multi-grained learning across the key components of network training—namely, the training data, network structure, and supervision strategy—thus limiting their performance. In this paper, we propose a multi-grained parallel spatio-temporal deepfake video detection architecture, which introduces a novel framework to mine more discriminative deepfake cues throughout the training pipeline. Firstly, we design a parallel spatio-temporal network combined with a cross-guided mechanism to concurrently extract frame-level spatial features and patch-level temporal features, while leveraging the relationship between spatial artifacts and temporal inconsistencies to enable multi-grained spatio-temporal synchronous learning. Secondly, we propose segment-level data augmentation strategies, including frame-random consistent self-blending and spatio-temporal data augmentation, which improve training data diversity at both frame and patch levels, thereby improving the model's ability to learn comprehensive deepfake representations. Finally, we construct a multi-grained supervision, comprising a patch-level temporal loss, a distance-based frame-level spatial loss, and a standard segment-level loss, for subtle deepfake feature learning. Extensive experiments demonstrate that our method possesses strong robustness and the generalization ability outperforms the current state-of-the-art methods across a series of deepfake datasets, including FaceForensics++, CelebDF, DFDC, DeeperForensics, and Faceshifter, on average.

Surgical tools detection for intraoperative surgical navigation system is essential for better coordination among surgical team in operating room. Because Orthopaedic surgery (OS) differs from laparoscopic, due to a large variety of surgical instruments and techniques making its procedures complicated. Compared to usual object detection in natural images, OS video images are confounded by inhomogeneous illumination; it is hard to directly apply existing studies that are developed for others. Additionally, acquiring Orthopaedic surgery videos is difficult due to recording of surgery videos in restricted surgical environment. Therefore, we propose a deep learning (DL) approach for surgery tools detection in OS videos by integrating knowledge of diverse representative surgery and non-surgery images of tools into the model using transfer learning (TL) and data augmentation. The proposed method has been evaluated for five surgical tools using knee surgery images following 10-fold cross validation. It shows, proposed model (mAP 62.46%) outperforms over conventional model (mAP 60%).

Deepfake technology refers to the use of deep learning to generate fabricated audiovisual content. It can be utilized in virtual reality and movie production. However, if used for malicious purposes, it can impact individuals and society and pose national security issues. Governments worldwide have introduced policies and legislation to address these issues. Detecting deepfake videos is thus critically important. This study analyzes methods of generating and detecting deepfake videos. We used frameworks such as DeepFaceLab and FakeApp to produce deepfake videos. Employing the Xception depthwise separable convolution network model as the foundation, we adopted the FakeVideoForensics method for continuous video analysis, the DeepFakes_FacialRegions method for specific facial feature analysis, and the Improved Xception method, which showed better performance. We have developed an enhanced depthwise separable convolution and facial feature extraction method for deepfake video detection, named Improved Xception Feature Fake Video Detection (IXFFVD). This method was trained using UADFV, FaceForensics++, Celeb-DF, and DFDC datasets, and it was used to detect face-swapped deepfake videos created by frameworks such as DeepFaceLab and FakeApp. Experimental results show that IXFFVD outperforms the compared methods. Although the detection efficacy of this method could be further enhanced, future work will focus on incorporating spatiotemporal detection criteria to improve the detection of deepfake videos created from different datasets.

We investigated whether people show signs of conflict detection in both more implicit and explicit judgments about the authenticity of short video clips depicting interviews with famous American actors. If so, they would be less confident when incorrectly seeing deepfakes as authentic than when correctly seeing authentic videos as authentic. Participants (N = 128; Mage = 33.7, SD = 12.1) from the USA were recruited on Prolific. Results showed that participants were more accurate at recognising deepfakes and less accurate at recognising authentic videos when they were explicitly asked to judge if a video was authentic or deepfake compared to more implicit authenticity judgments. Interestingly, they showed signs of conflict detection both when making more implicit and explicit authenticity judgments. These findings are relevant for the literature on both conflict detection in reasoning and decision-making and on deepfake recognition, as well as for research on training people to learn to recognise deepfakes.

The development of deep learning technology has allowed the emergence of the phenomenon of deepfakes, which is the manipulation of digital videos that resemble real videos with a high level of realism. These technologies pose serious threats to privacy, digital security, and the spread of false information. As the quality of deepfake videos increases, the detection of this fake content becomes increasingly challenging. This study aims to design and evaluate a deepfake video detection model using a combination of Convolutional Neural Network (CNN) and Gated Recurrent Unit (GRU). CNN is used to extract spatial features from each video frame, while GRU is used to capture the temporal relationships between frames. The dataset used is Celeb-DF(v2), which is a benchmark dataset that contains real videos and high-quality deepfake videos. The CNN-GRU model was trained and tested on the dataset, and its performance was evaluated using accuracy, precision, recall, and F1-score metrics.

Deepfake video detection has emerged as a critical challenge in the realm of artificial intelligence, given its implications for misinformation and digital security. This study evaluates the generalisation capabilities of the CoAtNet model—a hybrid convolution–transformer architecture—for deepfake detection across diverse datasets. Although CoAtNet has shown exceptional performance in several computer vision tasks, its potential for generalisation in cross-dataset scenarios remains underexplored. Thus, in this study, we explore CoAtNet’s generalisation ability by conducting an extensive series of experiments with a focus on discovering features and variations in deepfake videos. These experiments involve training the model using various input and processing configurations, followed by evaluating its performance on widely recognised public datasets. To the best of our knowledge, our proposed approach outperforms state-of-the-art models in terms of intra-dataset performance, with an AUC between 81.4% and 99.9%. Our model also achieves outstanding results in cross-dataset evaluations, with an AUC equal to 78%. This study demonstrates that CoAtNet achieves the best AUC for both intra-dataset and cross-dataset deepfake video detection, particularly on Celeb-DF, while also showing strong performance on DFDC.

The DeepFake video detection method based on convolutional neural networks has a poor performance in the dataset of hard compressed DeepFake video. And a large number of false tests will occur to the real data. To solve this problem, a networks model detection method for super-resolution reconstruction of DeepFake video is proposed. First of all, the face area of real data is processed by Gaussian blur, which is converted into negative data, and the real data and processing data are input into neural network for training. Then the residual network is used for super-resolution reconstruction of test data. Finally, the trained model is used to test the video after super-resolution reconstruction. Experiments show that the proposed method can reduce the false detection rate and improve the accuracy in detection of single frames.

FireMatch: A semi-supervised video fire detection network based on consistency and distribution alignment

- In recent months, the rise of free, deep learning- based software tools has significantly simplified the creation of highly realistic face-swapped videos, commonly known as "Deep Fakes" (DFs). While video manipulation through visual effects has been practiced for decades, recent advancements in deep learning have drastically enhanced the realism and accessibility of such synthetic content. Creating DFs using AI tools has become relatively straightforward; however, detecting these manipulations remains a major challenge. This is due to the complexity involved in training algorithms to recognize subtle and often imperceptible signs of tampering. In this work, we present a deep learning-based approach that combines Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) to detect Deep Fake videos. Our system employs CNNs to extract spatial features at the frame level, which are then passed to an RNN that captures temporal inconsistencies between frames introduced during Deep Fake generation. We evaluate our method on a large dataset of manipulated videos and demonstrate that our approach achieves competitive results using a relatively simple architecture. Key Words: Deep fake video detection, Convolutional neural networks(CNN), Recurrent neural networks(RNN)

Deepfake video detection methods, approaches, and challenges

The rapid advancement of generative adversarial networks (GANs) and other AI-driven synthesis techniques, deepfake videos have emerged as a significant threat to digital media integrity, enabling the creation of highly realistic but fake video content. These manipulated videos can be used maliciously in disinformation campaigns, identity theft, and other cybercrimes, making their detection a critical challenge. This paper presents a deep learning-based approach for deepfake video detection that leverages both spatial artifacts and temporal inconsistencies introduced during the manipulation process. The proposed method utilizes a hybrid architecture combining convolutional neural networks (CNNs) for frame-level feature extraction and recurrent neural networks (RNNs), specifically Long Short-Term Memory (LSTM) networks, to capture temporal dependencies across frames. We evaluate our method on widely used benchmark datasets including FaceForensics++, Celeb-DF, and DFDC, demonstrating superior accuracy, robustness, and generalization capability against multiple types of deepfake generation techniques. The experimental results validate the effectiveness of our approach in real-world scenarios, highlighting its potential for deployment in digital forensics, content moderation, and media verification systems. Keywords: Deepfake detection, Synthetic media, Deepfake forensics, AI-generated media,Face manipulation Video

The continuous growth of AI-generated synthetic media poses significant threats, including identity fraud, misinformation, and political manipulation, as deepfake video becomes increasingly indistinguishable from genuine recordings. The core problem is that these synthetic videos are often indistinguishable from authentic content to the human eye, rendering conventional forensic methods ineffective. This study therefore has the primary objective of developing a robust, intelligent detection mechanism capable of identifying subtle spatiotemporal artifacts in manipulated videos. The goal is to design and evaluate a hybrid deep learning architecture that leverages the complementary strengths of Convolutional Neural Networks (CNNs) for spatial feature extraction and Long Short-Term Memory (LSTM) networks for temporal sequence modeling, enhanced by an attention mechanism to focus on the most discriminative video segments. A key innovation of this work is the use of Spatiotemporal Chromatic Energy Distributions (SCED) as input features, which effectively model the harmonic relationships in authentic video and are sensitive to the inconsistencies introduced by synthetic generation processes. The results demonstrate the superior performance of the proposed hybrid CNN-LSTM model with attention, which achieved an accuracy of 98.7%, a precision of 96.2%, a recall of 94.4%, and an F1-score of 96.7%, significantly outperforming standalone CNN (88.5% accuracy) and LSTM (91.4% accuracy) baselines. In conclusion, this research confirms that the integration of SCED features with a hybrid CNN-LSTM architecture and attention mechanism provides a highly effective and reliable solution for deepfake video detection, offering a powerful tool to enhance the security and trustworthiness of digital media.

A novel model compression method based on joint distillation for deepfake video detection

Abstract: This research studies the use of Long Short-Term Memory (LSTM) networks for detecting DeepFake videos. DeepFake technology presents serious problems for preserving the authenticity and integrity of digital media since it uses sophisticated deep learning techniques to alter video content. We created a model capable of accurately detecting synthetic media by leveraging LSTM networks' sequential processing capabilities. The study focuses on taking advantage of temporal irregularities that are frequently found in DeepFake films but are difficult to identify using static analysis.

This exploratory study investigates the psychological and demographic factors influencing the ability to detect deepfake videos and examines the broader societal implications of deepfake proliferation.A comprehensive survey administered to 71 participants from two Midwestern universities collected data on psychological and cognitive factors, digital literacy levels, demographics, and deepfake detection abilities.The survey incorporated deepfake and genuine videos to assess participants' identification accuracy and confidence levels.Machine learning (ML) techniques were employed to analyze the data, revealing significant correlations between digital literacy, cognitive biases, and deepfake detection performance.The findings provide a foundation for understanding the complex interplay of psychological, demographic, and technological factors in the fight against deepfakes, emphasizing the importance of considering human dimensions in developing effective countermeasures.Therefore, they underscore the importance of enhancing public awareness and resilience against deepfakes to safeguard information integrity, personal privacy, and trust in digital media.