Smartphones are by far the world’s most popular cameras, but they certainly aren’t the best. Portrait photos, for example, often look awkward because of the wide-angle lens, flat because there is limited depth-of-field, and boring, because the photographer usually doesn’t have any special skills or access to advanced post-processing tools. Smartphone vendors including Google with the Pixel camera and Apple with the iPhone 7 Plus dual camera, are beginning to provide the needed hardware and some of the software to address these issues, but Adobe plans to push the envelope quite a bit further.
But will it make you a better photographer?
In a new video that teases the future of Adobe’s Sensei “AI” technology and its integration into smartphones, Adobe demonstrates how computational imaging can be used to address creating a more pleasing perspective and add a depth-of-field effect or even a different background. While the video is just a contrived video, it from watching it seems like these features would require some combination of either multiple imagers on the phone or the capture of several different versions of the portrait from slightly different perspectives — but amazingly Adobe says everything shown can be done with any portrait image in your camera roll.
Adobe takes things a step further in this video, and shows simple “one-click” access to pleasing photographic styles. This is similar to the style transfer capability that Adobe Research teased earlier this week, but simplified and shrunk into a smartphone platform. With current phones, advanced capabilities like transforming styles may require access to the cloud — which won’t make for the cool real-time previewing shown in the video. Longer-term, though, with the continuing advances in mobile GPU horsepower and smartphone storage capacity, eventually they’ll be able to run entirely on your mobile device.
Adobe’s Sensei is helping shape the future of image processing tools
Adobe is using Sensei as the marketing term for the combination of its massive content databases (everything available on Adobe Stock, for starters) and machine learning technology. Some of it, like the deep neural nets used for the style transfer research and image recognition capability, fit pretty well with the current definition of AI. Other pieces, like audience data analysis, include more traditional machine learning technologies. The sum total though, is a future with increasingly-powerful tools for our photography. It is also a future with more and more of the investment going into computational imaging on mobile devices, which doesn’t bode well for the market for traditional standalone cameras.