The relationship between artificial intelligence and photography just got real—Google’s latest AI experiment is taking Street View imagery from Google Maps and using post-processing techniques, operated by machine learning software, to create professional-grade landscape photography.
Completely devoid of a human touch, the software works by utilizing machine learning to train a deep neural network to scan thousands of Street View images. Then, after identifying which shots have notable landscape potential, the software “mimics the workflow of a professional photographer,” according to The Verge, and transforms that imagery into high-quality panoramas.
The project, led by Hui Fang, a software engineer on Google’s Machine Perception team, works by using one AI “photo editor” that identifies photographic elements, then tampers with them by cropping the photo, applying changes to lighting and color, and chooses a filter to apply. Next, another AI model compares the original shot to the edited image, resulting in software that understands the generalized “good” and “bad” qualities of a photograph. This process is called a “generative adversarial network,” a new AI technique that matches to AI networks against each other and uses the results to improve the system.
Finally, to test whether the AI software is actually producing professional-grade images, Fang and his team asked professional photographers to grade each image on a quality scale alongside side shots taken by humans. The Verge reports that “two out of every five photos received a score on par with that of a semi-pro or pro.”
This project will surely make some noise in the photography world, and really, it’s easy to see why: taking the human touch out of image-making is a concept that’s been confronted for some time now. As cameras and technology become increasingly highly-quality and user-friendly, the commoditization of the industry is a pervasive threat that’s growing each day. Photography is bigger now than ever and the tech industry wants to capitalize on the business.
However, it’s important to consider how AI software such as this could operate alongside your workflow. Imagine the time you would save if your software could scan large image sets to help make the best selects before beginning your post-work. Or even the ability to compare your edits to your raw images and have software shoot back some suggestions that may or may not create a more aesthetically pleasing final shot.
These days, being open to change is somewhat of a defining factor within the photography community, and projects like these are most definitely stirring the conversation.