AI has been the buzzword of this modern world for quite a long time. When we talk about words like ‘AI’, ‘bots’ or ‘Machine Learning’ many people start thinking about talking and walking robots — courtesy of our Sci-fi movies. But in reality, AI is nothing of this sort and has been with us for years now. It’s in your smartphones, mobile banking apps, car’s GPS system and even in the social media you use. It is in every field and photography is no exception. In fact, we are already using AI for photography in our smartphone’s cameras and you may not even know about it. Now, photography is no longer just a camera, lens, and a sensor — it is a collection of algorithms that immediately manipulate images to produce photographic results that would otherwise need hours of changes using some software tool. AI has been converting photography into a computational process. And this is just the beginning. In the next five to ten years, photography will be all about Artificial Intelligence and Machine learning.
AI and Photography
Nearly every photograph you see, would have been captured or created by a living person. No doubts, there are thousands of tools for creating images but most of them need a human presence to lead over the process. But if we talk about advancements — Microsoft has created an artist out of AI called “drawing bot.” The bot can create images from written descriptions of an object. This bot also sums some details to images that weren’t entered in the text, symbolizing that the AI has got a little imagination of its own.
It makes me think of a time in near future when you just have to feed some instructions in an app, such “I need a photo of mine standing in front of the statue of liberty” and voila it’ll be done.
Affect of AI on photographers
With more technological advancements in the future, AI tools may start to replace needed technical skills for photography. Google Clips is an AI-powered camera which can decide when the lighting or composition is aesthetically pleasing. This suggests that we are not long far from getting a totally automated photographer. Reviews of Google Clips haven’t been that good but the seed has been planted for sure. Also, with ever-evolving image-generation technology, businesses may be able to create their own images in the future, instead of hiring a professional photographer.
Generative Adversarial Networks & Photography
In words of Yann LeCun -
“GANs are the most interesting idea in the last 10 years in ML”
GANs are extensive neural net architectures composed of two nets, lying one against the other (“adversarial”). It is the foundation behind synthetic image generation. At the University of Montreal, Ian Goodfellow and other researchers introduced GANs in 2014 and since then it has been most widely used and fascinating aspects of Deep Learning. GANs can also be categorized as a set of generative models. Gans has a huge potential in photography and other fields as it can learn to mimic any set of data. Gans can be defined as a robot artist and output has been good till date. A portrait generated by GANs was sold for $432,000. Uses of this technology can be seen in Inpainting and Outpainting, face synthesis and GANimation.
Painting to Photograph Translation
GANs can also be used to make photographs more realistic or simply turn a painting into a new photograph. For this process a different type of GAN is used, know as CycleGAN. This uses two discriminators and two generators.
What do we expect from here?
AI is sure to have a huge impact on photography and imaging in the near future. Technologies like GAN are capable of generating photographs from text inputs. You can’t deny the fact, in foreseeable future such technologies will be able to generate high-resolution photographs and videos with simple commands. Think about a whole video created by feeding some codes in GAN, isn’t it great? In the future, there might be no need of tools like CorelDRAW and Photoshop to enhance the images. The only thing that haunts me is — will it be end of real photography?