Back to insights

The positive uses of the technology behind deep fakes

The development of Generative Adversarial Networks (GANs) has meant that artificial intelligence can now be used to create fake images based on combining and superimposing existing images onto source images or videos. This production of “Deep Fakes” has received some bad press in recent years but the use of this technology doesn’t have to always be used in a negative way.

  • Helping us to choose a new wardrobe
  • Showing us how a new hairstyle looks
  • Provide insight into furnishing our homes
  • Assist in improving image quality
  • Training medical professionals and contributing to medical research
  • Data sharing
  • Cybersecurity
  • 3D object generation
  • Text to image synthesis

What are GANs, and why are there concerns about them?

Having been given the name “Deep Fakes” it is obvious the stigma and suspicion that has aroused around the use of Generative Adversarial Networks. The negative connotations and fear surrounding GANs are unsurprising given how they have been used to create fake news as well as sinister hoaxes. There have also been cases where deep fakes have been used to create fake pornographic videos with the face of a celebrity imposed on the actors.

However, while these concerns are understandable, and the perpetrators of such crimes should be held to account, the use of this artificial intelligence should not be discounted entirely. There are many benefits to have from the use of GANs, and they have the potential to have a highly positive impact on the decisions we make when it comes to purchasing or getting that perfect image.

How can GANs be used positively?

Buying clothes online

Shopping online has revolutionized the way we buy clothes thanks to the vast variety of choice, as well as the ease with which we can purchase a whole new wardrobe from the comfort of our front room. However, nothing beats the opportunity to try something on and with around half of online clothing purchases being returned, it is clear that it is hard to tell the right fit of an item from a photo.

With the use of AI and GANs, developers are now able to generate an image of a human figure of any size, body shape, and skin colour and predict how that particular clothing item would fit on a person of that physique. You can even add accessories to the image to see if your favourite type of shoes or handbag would look good with the outfit choice.

Try new hairstyles

Taking the step to go for the chop or try out a bold new hair colour can be a big decision. Also, many of us stick with the same old style for fear of what a new hairstyle might look like as once that decision has been made, we can’t go back.

Well, with the assistance of GANs, it is now possible to reconstruct a photo of yourself to see how you would look with a different characteristic. Whether it be a short bob, hair extensions, dreadlocks or bright pink hair, GANs can translate a part of one image onto another image so that you can see yourself differently.

This incredibly smart technology doesn’t stop there; it is also possible to use GANs to see how we are going to look as we age. That’s right; artificial intelligence can now even show us how we will look in 50 years — preparing us for what is yet to come!

Furnish your home with the use of GANs

Even with a visit to the furniture store, it is still tricky to envisage how a sofa, vase or even a light fitting will look in our home. With GANs it is possible to furniture match — take a photo of your dining room and find a vase that would fit the colour of the room. Having the ability to synthesize images with a high degree of realism, GANs have the potential to allow us to see how new furniture will look in our homes before we decide to purchase them.

This use of GANs has already been explored by estate agents who have added furniture to empty spaces to make homes look more appealing to potential buyers. The next step is allowing consumers to see how specific items will look in their homes and we expect this technology to be put use by online furniture stores as this form of artificial intelligence continues to be developed.

Repair and amend images

Ever tried to get a photo of your child for a passport but they can’t help but smile? No problem, you can change that smile to a straight face with the assistance of GANs. The use of GANs in image enhancement and alteration is growing, and if your a keen photographer or are interested in digital images, this use of GANs will be right up your street.

In addition to being able to create new images, GANs can also repair images, enhance image quality, fill in missing pieces and remove objects from images, remove imperfections, and even alter facial expressions. Other recent developments have seen the use of GANs in reconstructing a 3D model from a single 2D image.

Training health professionals

Another of the many benefits of using GANs is it’s potential to generate medical images. In the past training has been inhibited by a lack of patient data (or the absence consent to use such data) and a lack of consistency in terms of access to medical images across the board. With the help of GANs, however, the healthcare sector is fast seeing the potential to create artificial data for use in medical training.

For example, GANs have been used to create fake MRIs of brain tumours for doctors being trained in tumour segmentation. Images generated through the use of GANs have also been used in other medical training such as tissue recognition, liver lesion classification, and retinal observations.

While in it’s early stages, the use of GANs in medical research and training has the potential to be significant. Where data is lacking or inconsistent images used by GANs are already being used, albeit with the requirement of a big budget, the right talent and time to create them.

Other positive uses of GANs

Data Sharing: for many companies, their data needs to be kept private, and sharing be kept to a minimum. However, there do come occasions where it is beneficial to share data with a third party (perhaps a researcher, consultant or volunteer that has come to work at the organization). In these cases, with the use of GANs, an example could be generated from the original data which does not contain anything confidential rather, it gives an idea of what data the company is dealing with.

Cyber Security: the nature of GANs means that new data is generated and then it is assessed by the discriminator (who is responsible for telling the difference between fake data and a genuine example). Each role improves over time as they learn from mistakes and build on experience. This means that deep learning models become more attuned and experienced in identifying potential cyber attacks.

3D Object Generation: all gamers out there will be pleased to hear that GANs can also assist in creating 3D characters in your favourite computer games. Through automating this process, GANs will be able to save countless hours of development, which will surely result in an abundance of new characters.

Text to image synthesis: translating text descriptions into an image has long been a challenge to overcome for developers. Here is another field where the use of GANs is making significant steps in a formerly underdeveloped field. Watch this space.

The use of GANs as “deep fakes” has been the overwhelming rhetoric in response to this new use of artificial intelligence. And while these concerns are not unfounded, the positive uses of GANs should not be discounted. From transforming the way we buy our clothes, assisting with a new hairstyle, improving our images and even contributing to medical research and training we believe that the incredible potential of GANs will soon be difficult to ignore.