In recent years, all of us have grown quite comfortable with an AI completing our sentences because it makes our work a lot easier and effective. An example of this would be our very own emails wherein the AI has been a life saver almost all the time. BUT, that is old news now. So what's next? Well, here is the next big thing IMAGE GPT.
What exactly is IMAGE GPT?
So, in the above images , the first column contains the image that was fed to the AI and the next four images was what the AI predicted ,while the fifth column contains the original image. So, this is exactly what IMAGE GPT does.
It is an AI model that can create almost realistic and sometimes even better than the original image from scratch. It can create images that it has never seen before.
How does it do it?
It uses a machine learning technique called Generative Pretraining. In this the AI model is taught to create a new image on its own by giving enormous datasets as examples that is it is trained with lots(say billions) of images wherein it is able to understand how an image would look like without seeing it previously.
Now, the AI model is initialized with pixels which it uses to predict the next pixel and so on until the entire image is predicted. It uses two techniques to predict which are Auto Regression and Attention.
Auto Regression is how the AI predicts the next value in sequence and Attention is a technique that allows the AI model to focus on certain specific and important things of the input sequence.
Where is it actually used?
It has its uses in a variety of things especially Virtual Reality(VR), Fashion, Photography etc. In VR it can be used to create the environments of games , in Photography it can be used to remove unwanted objects from a real picture and in Fashion it can be used to create new and trendy designs.
What does it lack?
Although it is a revolutionary concept it has its own limitations. Even after training it with billions of data it hasn't reached the human intelligence level and hence can sometimes be illogical. The computational costs are very large so the resolutions that are given to the model are low and hence the outcome is affected.
Conclusion
Well, IMAGE GPT is an AI model that is used to generate images from the scratch that are quite realistic and that can be used in a variety of fields. Despite of its limitations it wouldn't be wrong to say that it is the next revolution in technology and the upcoming models would only be more sophisticated.