Interest in the phenomenon of “deepfakes” has increased in recent months, presumably as the public comes to terms that people can and use #AI to create super-realistic fake videos and images. Where as Deep learning also known as deep structured learning is part of a broader family of machine learning methods based on learning data representations, as opposed to task-specific algorithms. With that said I want to ignite a discussion on what deep fake is and how can we use it in something thats useful to human-being?
DeepFakes offers the ability to swap one face for another in an image or a video. Face swapping has been done in films for years, but it required skilled video editors and CGI experts to spend many hours to achieve decent results.
Example, as shown in attachment.
Interesting topic! In my discussion I am going to focus more on the first part of your question about what deepfakes are, and I will try to mention some interesting works in the field. I am not a big fan of using the term "fake" to refer to the images/videos generated by a deep network. Instead I prefer the term "synthetic", since the images/videos are essentially simulated.
Deepfake is a term coined to refer to synthetic images (frames incase of videos) that are created by a deep learning network specifically a generative adversarial network (GAN). GANs were first introduced by Ian Goodfellow for unsupervised learning, and over time several types of GANs were developed for various synthesis/translation tasks. Reliable image synthesis was first possible with DC-GAN, and later SA-GAN and PG-GAN were able to generate high-resolution images of 1024x1024!
Today, GANs are capable of generating highly realistic images that are very convincing to human brain. One of the recent work that caught my attention (arxiv link to the paper: Article Synthesizing Images of Humans in Unseen Poses
) is a method developed by researchers from MIT, that is capable of translating a person's pose from doing one action to another. For example, if you input an image of a person playing tennis, the method could output a synthetic image of the same person doing yoga or playing golf! The authors were also able to generate synthetic videos of a person doing a task that is different from what the person was doing in the input video. From a security standpoint, I think that's concerning. Imagine that you recorded a video of you shaking hands with your friend, it could be easily translated to a video of you punching your friend! That being said, such methods could also be beneficial in many different ways which I will briefly discuss later.As opposed to translating an image or video of a person, some methods focus on generating an image of a completely new person. There is an interesting website https://thispersondoesnotexist.com/ - everytime you reload the webpage, it generates an image of a person that does not exist - a deepfake image. The GAN behind this work is a StyleGAN which was introduced by a group in NVIDIA - Karras is the lead author, and he was also the lead author in the development of PG-GAN. StyleGAN is the current state-of-the-art in style based image synthesis network.
Here's the arxiv link to the paper: Preprint A Style-Based Generator Architecture for Generative Adversar...
. The purpose of a StyleGAN is to not only learn to synthesize an image, but also to synthesize variations in specific attributes of the image. For example, in a human face the StyleGAN would learn different hair patterns, skin textures such as freckles etc., different poses and views, and other attributes. Because this type of GAN is capable of controlling these attributes when synthesizing an image, a wide variety of images that look convincingly realistic can be generated.Now that we have the AI technology capable of generating realistic fake images, some obvious questions pop in our head: Why is it useful? What are the shortcomings? Are they dangerous? Where are we headed with the deepfake technology? etc. etc.
Although these methods have been raising concerns on safety and security such as generating fake news videos, I think that there are several benefits of them. For example, the network capable of generating fake people can be used in criminal investigations to generate different variations of a person's face with minimum description provided by the witness. They could be used to simulate crime scenes, for example, synthesizing different poses of the suspect at the crime scene could likely reveal information about how the crime might have happened.
Another application I could think of is gaming and entertainment. The synthesized images and videos can be used for hyperrealistic game design. Today, there is a huge market for games with high quality graphics, and GANs can be really helpful in enhancing the industry.
Synthetic images could be really helpful in predicting future events. For example, a GAN could be trained to generate future video frames of a moving vehicle and estimate if the vehicle would collide or not. I think there was an article where the authors developed a prediction GAN, but I haven't followed up on this trend recently.
GANs are immensely powerful in changing the future when used in the right way, and I am looking forward to a lot more applications of synthesizing images and videos especially in transforming healthcare.
DeepFakes offers the ability to swap one face for another in an image or a video. Face swapping has been done in films for years, but it required skilled video editors and CGI experts to spend many hours to achieve decent results.
Example, as shown in attachment.
Its become another challenge for researcher in various of fields, if deepfake able to produce something fake, that means the challenge to detect whether its a fake or real. a standard procedure for fake detection should be discussed either to form counter attack againts deep fake. because if its too smart its might be become a new threat to the society.
@Ahmad Hoirul Basori
I agree with the countering of synthetic images in all. As GANs become better at creating these synthetic images and videos not even regulations will be able to keep up. It will be quite difficult to tell the true apart. We need to think of ways to counter as opposed to control.
Dear Mohammed Kemal,
Deepfake (a portmanteau of "deep learning" and "fake" is a technique for human image synthesis based on artificial intelligence. It is used to combine and superimpose existing images and videos onto source images or videos using a machine learning technique called a "generative adversarial network. The combination of the existing and source videos results in a video that can depict a person or persons saying things or performing actions that never occurred in reality. Such fake videos can be created to, for example, show a person performing sexual acts they never took part in, or can be used to alter the words or gestures a politician uses to make it look like that person said something they never did.
Because of these capabilities, deepfakes have been used to create fake celebrity pornographic videos or revenge porn Deepfakes can also be used to create fake news and malicious hoaxes.
https://en.wikipedia.org/wiki/Deepfake
Regards,
Shafagat
A Deepfake is an AI-assisted video created by taking a number (usually hundreds or thousands) of photos of a source person. These images can be downloaded from a number of sources, such as the person’s Instagram, Facebook, Snapchat, or even a Google image search).
The Deepfake AI software maps the face of the source images and generates a 3-D face model based on the photos it is fed. The software is also given a source video, containing a target face which the user wants to replace. The AI also maps the face of the person in the video, again creating a 3-D mapped model.
Using a machine learning technique called a "generative adversarial network" (GAN) is a new method of creating fake video images. Notably, deepfakes are created through AI and do not require the essential skills that it would take to create realistic videos. That means that just about anyone could create a deepfake to produce their own intended images. One of the caveats in producing deepfakes is that people will take such videos at face value and will stop trusting the validity of any video content at all. The truth is that all of reality ,as we once knew, seems to be doomed because of dramatic advances in image creating technologies.
Deepfake is an artificial intelligence based technology used to produce or alter video content so that it presents something that didn't, in fact, occur. Please take a look at the following RG links for more details.
Preprint DeepFakes: a New Threat to Face Recognition? Assessment and Detection
Article Artificial Intelligence, Deepfakes and a Future of Ectypes
Conference Paper Detection of Deepfake Video Manipulation
A deepfake is a video of someone saying and maybe even doing things that they never did in reality. These fake videos are created on computers using AI technology that allows the creator to use existing footage of an individual.
Deepfake is very well explained in youtube videos and people around the world are getting aware of this technology which is used to create the mockery of the reality. The scope of human suffering is broaden in all dimensions in the form of heightened psychological pains, frustrations , internal and external conflicts, aggression, child abuse and many other of the same type.The reasons are quiet varied and complex but the major reason that have emerged in recent times is related with the role of highly polluted self image that is the result of opening oneself significantly with fantasies and watching the digital version of the same for temporary pleasures.The contamination of consciousness with act result in bitter personal,social and emotional consequences and deepfake technology has the capacity to become an emotional and social hazard with server consequences of unparalleled nature.On the contrary there are some benefits in the form of producing unmatched recreational ventures, a use maybe in psychotherapy, simulations and innovative uses in education,research and other fields of knowledge.The unattended use of this technology can be source of social and emotional havoc.
Arvind Singh
Thank you for sharing the link. This really answers most of the concern that DeepFakes has. There is no problem with the technology but the real problem is how people is using it.Preprint DeepFakes: a New Threat to Face Recognition? Assessment and Detection
great discussion, thank you all: Mohammed Kemal Olutosin Ademola Otekunrin Nirmala S.V.S.G
Muhammad Zubair
- Badges
- Science topic
- Similar topics
- Psychology
- Mental Processes
- Learning