GAN is a friend or foe?

Tariq, Shahroz; Lee, Sang Yup; Kim, Ho-Young; Shin, Youjin; Woo, Simon S.

doi:10.1145/3297280.3297410

Cited by 44 publications

(5 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…A faceswap replaces the facial features of target person's with those of source person, while expression swap modified the facial features of target person based on those of source person. Quite some work has been done to detect deepfakes [10,14,33,13,34]. Many of these techniques employ machine learning and deep learning for image forensics based on the pixel-based attributes.…”

Section: Deepfake Image Detectionmentioning

confidence: 99%

“…Dang et al [37] identified the GAN images by employing a CNN that works on the pre-processed dataset (only tampered or GAN region of the image) to recognize the fake image. On the other hand, Tariq et al [13] devised the classifier for human created and Progressive GAN images and have also devised the end-to-end framework for automatic detection [34].…”

Section: Deepfake Image Detectionmentioning

confidence: 99%

See 1 more Smart Citation

Exposing AI Generated Deepfake Images Using Siamese Network with Triplet Loss

Kanwal

Tehsin

Saif

2022

cai

View full text Add to dashboard Cite

Generative Adversarial Networks have gained popularity mainly due to their ability to create fake human faces. The remarkable detail with which such images have been created in the past few years has exceeded the ability of humans to differentiate between these fake images and real images. Such images have been known to be capable of deceiving the face recognition systems with certain success as well. Forensic systems being developed nowadays take into account adversarial attacks in order to create a more comprehensive detection approaches. Different GAN algorithms such as StackGAN, StyleGAN use different architectures to produce images. Since the underlying technique is different from one another it is difficult for any single detection algorithm trained on one kind of GAN to detect fake images generated from some other kind of GAN. In this research we use a siamese network with triplet loss function to provide a generic solution for detection of GAN generated images or deepfake images. Extensive experiments have been conducted to analyze the effectiveness of the proposed approach. The results show that the siamese triplet loss network performs significantly better than the contemporary approaches with accuracy exceeding 90 % in most experiments.

show abstract

Section: Deepfake Image Detectionmentioning

confidence: 99%

Section: Deepfake Image Detectionmentioning

confidence: 99%

Exposing AI Generated Deepfake Images Using Siamese Network with Triplet Loss

Kanwal

Tehsin

Saif

2022

cai

View full text Add to dashboard Cite

show abstract

“…Kuwahara et al 48 study the applicability of statistical anomaly detection methods to identify malicious CAN messages, where a pipeline technology is proposed to extract the timestamp and ID information in each messages quickly, and the efficiency of the proposed method is evaluated in real message datasets and in supervised and unsupervised cases. Besides NNs, other artificial intelligence algorithms, such as LSTM, 49 Bayesian networks, 50 hidden Markov models, 51 SVM, 52 compound classifier, 53 and singular spectrum analysis 54 are also introduced to build an IDS for CAN bus.…”

Section: The State‐of‐the‐art Work About Security Protection Of In‐vehicle Networkmentioning

confidence: 99%

Cybersecurity protection on in‐vehicle networks for distributed automotive cyber‐physical systems: State‐of‐the‐art and future challenges

Xie

Zhou

et al. 2021

Softw Pract Exp

View full text Add to dashboard Cite

The ever‐evolving trip mode of human being leads the automobiles moving toward connected, autonomous, sharing, and electrified vehicles rapidly. But the connection introduces new cybersecurity problems on in‐vehicle networks, which poses great challenges for safety guarantee of distributed automotive cyber‐physical systems. This article first analyzes the cybersecurity vulnerabilities and defines the security requirements for in‐vehicle networks, and then introduces the architecture evolution of in‐vehicle network. Based on the definition on architecture of in‐vehicle networks, this article defines a security protection framework for it. And then, it surveys the state‐of‐the‐art works for availability protection, integrity protection, and confidentiality protection of in‐vehicle networks, respectively, and detailed analysis and comparisons are given about the proposed cybersecurity protection mechanisms. Finally, it summarizes the future challenges for cybersecurity protection of in‐vehicle networks, and proposes possible solutions for these challenges.

show abstract

“…Desarrollo de una aplicación web que genere rostros de personas que no existen en el mundo real En el campo de la inteligencia artificial, las Redes Generativas Adversarias o Generative Adversarial Networks (GANs) han emergido como una poderosa herramienta que desafía las fronteras de la creatividad digital (Tariq et al, 2019). Estas redes, conceptualizadas por primera vez por Ian Goodfellow y sus colegas en 2014, han revolucionado la generación de contenido visual, ofreciendo la capacidad única de crear imágenes realistas e inéditas que desafían la realidad misma.…”

Section: Introductionunclassified

Desarrollo de una aplicación web que genere rostros de personas que no existen en el mundo real

Quezada-García,

Cajamarca-Castillo,

Rivas-Asanza

et al. 2024

IBEROTECS

View full text Add to dashboard Cite

Con el creciente desarrollo de la Generative Adversarial Network (GAN), la generación de imágenes son un desafío emocionante en el campo del aprendizaje profundo y la inteligencia artificial. A nivel internacional se han desarrollado diversos trabajos relacionados a modelos de redes generativas antagónicas, pero una brecha significativa persiste en la falta de comparativas entre diferentes algoritmos. La capacidad de generar imágenes inexistentes que se asemejen en gran medida a las imágenes del mundo real es interesante para muchos casos de uso. En la presente investigación se propone desarrollar una aplicación web basada en redes neuronales preexistentes, utilizando la tecnología generativa antagónica para generar rostros de personas. En este estudio, se consideró como población de estudio a los 5.000 rostros que pertenecen el conjunto de datos FFHQ Face Dataset (2.500 rostro mujer) y (2.500 rostro de hombre). Las técnicas que fueron empleadas están basadas en dos modelos utilizando los siguientes algoritmos: Generative Adversarial Network (GAN) y Red Adversarial Generativa Convolucional Profunda (DCGAN). Las mediciones estadísticas de la matriz de confusión como resultado de la clasificación se utilizaron como métricas de rendimiento. Los resultados permiten concluir que el modelo (DCGAN) es el mejor método de clasificación, debido a su mejor predicción de los valores obtenidos de verdaderos positivos, y verdaderos negativos, además, la media de la precisión del modelo fue de 85,02%, superior al otro modelo. La aplicación desarrollada ha cumplido con éxito el objetivo principal de generar rostros humanos a partir de algoritmos de inteligencia artificial.

show abstract

GAN is a friend or foe?

Cited by 44 publications

References 17 publications

Exposing AI Generated Deepfake Images Using Siamese Network with Triplet Loss

Exposing AI Generated Deepfake Images Using Siamese Network with Triplet Loss

Cybersecurity protection on in‐vehicle networks for distributed automotive cyber‐physical systems: State‐of‐the‐art and future challenges

Desarrollo de una aplicación web que genere rostros de personas que no existen en el mundo real

Contact Info

Product

Resources

About