AI Turns Photographer in Google Street View


No artificial intelligence photographer is walking on our streets yet, but one sure is walking on Google’s.

There’s a good reason we don’t have robot photographers yet, the technology hasn’t yet caught up to the idea. But before it does we can play around within our already established digital environments and provide artificial intelligence with an imaginary tripod and a virtual, algorithmically based, camera lens. And Google, a big-time player in the field of deep learning and the creator of DeepMind, did just this when they used a basic set of instructions for taking a photograph in Google Street View and gave it to a “deep-learning system for artistic content creation” they call Creatism.

A total of around 40,000 spherical panorama images, originally grabbed by their global armada of camera trucks, were given to the artificial photographer. It was then told to do four things. First, it needed to crop a section of the original photograph to get an editing sample. It then moved on to apply two filters, first one for saturation and then one for High-Dynamic Range (HDR). Finally, it applied a number of filter masks using Snapseed, the company’s lightweight photo-editing application for smartphones.

This shows the step-by-step process, from cropping to masking. PHOTO CREDIT: Google

To evaluate how close the resulting photographs were to a professional level they gathered six human professional photographers. The judges could give a score from one to four, with their underlying meaning as written below.

“1: Point-and-shoot without consideration for composition, lighting etc.
2: Good photos from general population without a background in photography. Nothing artistic stands out.
3: Semi-pro. Great photos showing clear artistic aspects. The photographer is on the right track of becoming a professional.
4: Pro.”

Noteworthy is that almost 40 percent of all the photographs were rated as “semi-pro” or “pro” level. That’s a pretty decent score, to say the least. Arguably, the most impressive about this feat is that it manages to use a spherical photograph, analyze it, and crop it in a way that provides such a convincingly professional composition. Looking at some of these photographs, it’s certainly hard to argue against their aesthetic appeal.

Google says that these panoramas were a testing bed for a larger project. In the future, they may be able to use this type of technology to help people grab better images themselves, but what form such digital assistance may take, and when it might happen, is discussed in neither the research papers nor their recent blog entry.