AI Creates Near Perfect Images Of People, Dogs and More

Dear fellow scholars, this is two minute papers with in the last few years we have seen a bunch of new AI based techniques that were specialized in generating new and novel images. This is mainly done through learning based techniques, typically a generative adversarial network, a gan, in short, which is an architecture where a generator neural network creates new images and passes it to a discriminator network which learns to distinguish real photos from these fake generated images. These two networks learn and improve together, so much so that many of these techniques have become so realistic that we often can't tell they are synthetic images unless we look really closely. You see some examples here from Biggan, a previous technique that is based on this architecture. So in these days many of us are is there life beyond gans? Can they be matched in terms of visual quality by a different kind of technique? Well, have a look at this paper because it proposes a much simpler architecture that is able to generate convincing high resolution images quickly for a ton of different object classes. The results it is able to churn out is nothing short of amazing. Just look at that. To be able to proceed to the key idea here, we first have to talk about latent spaces. You can think of a latent space as a compressed representation that tries to capture the essence of the dataset that we have at hand. You can see a similar latent space method in action here that captures the key features that set different kinds of fonts apart and presents these options on a 2D plane. And here you see our technique that builds a latent space for modeling a wide range of photorealistic material models. And now onto the promised key idea. As you have guessed, this new technique uses a latent space, which means that instead of thinking in pixels, it thinks more in terms of these features that commonly appear in natural photos, which also makes the generation of these Images up to 30 times quicker, which is super useful, especially in the case of larger images. While we are at that it can rapidly generate new images with the size of approximately a thousand by thousand pixels. Machine learning is a research field that is enjoying a great deal of popularity these days, which also means that so many papers appear every day, it's getting finding difficult to keep track of all of them. The complexity of the average technique is also increasing rapidly over time. And what I like most about this paper is that it shows us that surprisingly simple ideas can still lead to breakthroughs. What a time to be alive. Make sure to have a look at the paper in the description as it describes how this method is able to generate more diverse images than previous techniques and how we can measure diversity at all, because that is no trivial matter. This episode has been supported by Weights and Biases. Weights and Biases provides tools to track your experiments in your deep learning projects. It is like a shared logbook for your team and with this you can compare your own experiment results, put them next to what your colleagues did and you can discuss your successes and failures much easier. It takes less than five minutes to set up and is being used by OpenAI, Toyota Research, Stanford and Berkeley. In fact, it is so easy to add to your project. The CEO himself, Lucas, instrumented it for you for this paper and if you look here you can see how the output images and the reconstruction error evolve over time and you can even add your own visualizations. It is a sight to behold really, so make sure to check it out in the video description and if you liked it, Visit them through wandb.com papers wandb. Com papers or just use the link in the video description and sign up for a free demo today. Our thanks to Weights and Biases for helping us make better videos for you. Thanks for watching and for your generous support and I'll see you next time.