Google has brought Street View to some extraordinary places across the planet. We can cruise down rural roads and up hiking trails, stroll through village streets and wander down winding mountain paths all from the comfort of our couches. It is educational and entertaining to wander the world with Street View taking in the sights, but what if those sights could be on par with images made by professional photographers? That's what Google is exploring using artificial intelligence.
The technical challenge is especially interesting because there isn't a right or wrong answer for transforming an image into art. It takes subjectivity, something still being explored in machine learning (ML).
Google Research Blog writes, "To explore how ML can learn subjective concepts, we introduce an experimental deep-learning system for artistic content creation. It mimics the workflow of a professional photographer, roaming landscape panoramas from Google Street View and searching for the best composition, then carrying out various postprocessing operations to create an aesthetically pleasing image."
Google's machine was "taught" what makes a photo beautiful through showing it professional photographs and modifying them with filters that degrade the quality of the images to make them look "bad" — all of which helped the machine learn the exposure, saturation and other details that make a photo look beautiful.
The "virtual photographer" explored some 40,000 panoramas from Street View treks spanning the Alps, Big Sur in California, Yellowstone National Park and other locations. It then selected compositions, processed the images and created photographs. About 40 percent of professional photographers asked to rate the quality of the images judged them to be at the "semi-pro" or "pro" level.
Now, you be the judge. Here's Google's machine improving a view of Parco delle Orobie Bergamasche in Italy. What do you think?
And here is a landscape from Interlaken, Switzerland: