“There’s a bunch of ways we can select the ‘best’ photos,” says
Google Photos product manager Francois de Halleux. “We use a lot of machine learning to detect the elements in a photo that make it of better quality than another. We also eliminate duplicates.”
In many cases, the system pegs photos with landmarks in them as the “best” shots. And it can identify those famous places without using your location data at all, although geotagging does help with some fact-checking. In the most-useful cases, it will identify and provide names of landmarks—helping you remember where you went or simply spell things correctly.
“We can detect landmarks, we have 255,000 landmarks that we automatically recognize,” says de Halleux. “It’s a combination of both computer vision and geotags. Even without the geotags, we’d be able to recognize a landmark.”
In those cases, Google’s system would recognize a landmark, then double-check that recognition against a location or the photo’s geotagging. In some cases, it helps the system identify the real deal versus a replica.
“If we see a photo of the Eiffel Tower, we know the person is in Paris… or in Las Vegas,” explains Google Photos product lead David Lieb.
Vacations aren’t the only use case for the new Assistant-created albums, but they’re likely to be the most common scenario for the auto-generated albums. Lieb and de Halleux say that the Assistant looks at the distance you are away from your home as a trigger for album auto-creation, but it also pays attention to how many pictures you’ve taken in a short period of time and whether it’s a national holiday or other significant day.