[Paper Review] Deepfakes Generation and Detection: State-of-the-art, open challenges, countermeasures, and way forward

Momina Masood, Marriam Nawaz+3 more|arXiv (Cornell University)|Feb 25, 2021

Digital Media Forensic Detection4 citations

TL;DR

This paper provides a comprehensive survey of state-of-the-art deepfake generation and detection techniques, focusing on both visual and audio modalities using deep learning, particularly GANs. It analyzes existing tools, datasets, evaluation standards, open challenges, and countermeasures, offering a roadmap for future research in detecting and mitigating synthetic media threats.

ABSTRACT

Easy access to audio-visual content on social media, combined with the availability of modern tools such as Tensorflow or Keras, open-source trained models, and economical computing infrastructure, and the rapid evolution of deep-learning (DL) methods, especially Generative Adversarial Networks (GAN), have made it possible to generate deepfakes to disseminate disinformation, revenge porn, financial frauds, hoaxes, and to disrupt government functioning. The existing surveys have mainly focused on the detection of deepfake images and videos. This paper provides a comprehensive review and detailed analysis of existing tools and machine learning (ML) based approaches for deepfake generation and the methodologies used to detect such manipulations for both audio and visual deepfakes. For each category of deepfake, we discuss information related to manipulation approaches, current public datasets, and key standards for the performance evaluation of deepfake detection techniques along with their results. Additionally, we also discuss open challenges and enumerate future directions to guide future researchers on issues that need to be considered to improve the domains of both deepfake generation and detection. This work is expected to assist the readers in understanding the creation and detection mechanisms of deepfakes, along with their current limitations and future direction.

Motivation & Objective

• To provide a systematic review of deepfake generation and detection methodologies across visual and audio modalities.
• To analyze current public datasets, evaluation standards, and performance metrics used in deepfake detection research.
• To identify open challenges and limitations in existing deepfake detection systems.
• To propose actionable future research directions for improving deepfake detection and generation technologies.
• To support researchers and practitioners in understanding the technical and ethical implications of synthetic media.

Proposed method

• Surveying and categorizing existing deepfake generation techniques, with a focus on Generative Adversarial Networks (GANs) and other deep learning architectures.
• Analyzing machine learning-based detection methods for both visual and audio deepfakes, including convolutional neural networks and recurrent models.
• Evaluating performance using standardized benchmarks and public datasets such as FaceForensics++, Celeb-DF, and VGGSound.
• Comparing detection accuracy, F1-scores, and AUC values across different models and datasets to assess robustness.
• Identifying common weaknesses in detection models, such as generalization failures on unseen datasets and adversarial robustness issues.
• Proposing a structured framework for future research based on identified gaps in data, evaluation, and model generalization.

Experimental results

Research questions

• RQ1 What are the dominant deep learning architectures used in modern deepfake generation, particularly in visual and audio domains?
• RQ2 How do current detection models perform across different public datasets, and what are their key limitations in real-world deployment?
• RQ3 What are the major open challenges in deepfake detection, including generalization, robustness, and data scarcity?
• RQ4 What countermeasures and standards are currently in place for evaluating and improving deepfake detection systems?
• RQ5 What future research directions are most critical for advancing both deepfake generation and detection technologies?

Key findings

• Current deepfake detection models show high performance on in-domain datasets but often fail to generalize to out-of-domain or unseen manipulation techniques.
• The FaceForensics++ and Celeb-DF datasets are widely used benchmarks, with top detection models achieving AUC scores above 0.95 on these datasets.
• Audio deepfake detection remains underexplored compared to visual deepfakes, with fewer standardized datasets and evaluation protocols.
• Many detection models are vulnerable to adversarial attacks, indicating a need for robustness improvements.
• A lack of standardized evaluation metrics and diverse, real-world data limits the practical deployment of current detection systems.
• Future research must prioritize generalization, interpretability, and the development of unified benchmarks for both visual and audio deepfakes.