Google Releases New Camera Trained By Real Photographers
Perhaps the strangest of the new products launched last year is ‘Clips’, Google’s automatic camera.
Now available for pre-order (edit: it appears to be sold out, but you can join the waiting list), Google Clips is a wireless 50mm-square camera designed to capture still images and ‘motion photos’ (short video clips without sound) without direction from a human photographer.
Instead, Clips uses machine learning to determine automatically what makes a good photo or video clip. The camera is able to detect ‘great expressions, lighting and framing’ and is continuously learning. It can, for example, learn to recognise the faces of the people you know and focus on them in preference to others.
The idea is to point Clips at a potentially interesting scene, such as a family get-together, and leave it to snap away, automatically capturing the most interesting and memorable moments.
Google’s Josh Lovejoy provides more detailed information on how this is achieved in a recent Google Design blog post, including how human experts rather than machines are the driving force behind Clip’s AI-based skills.
Lovejoy set out to interview and hire expert photographers to evaluate footage captured by people on his team. The goal was to answer the question, “What makes a memorable moment?”. Interviewees included a documentary filmmaker, a photojournalist and a fine arts photographer.
It turns out that one of the most important steps was to take exactly the opposite approach. By teaching the AI instead what a bad image looks like, the camera wouldn’t then waste time processing images which were clearly useless.
A ‘useless’ image could be something like a capture made accidentally inside a pocket or bag or perhaps when an object such as a finger strays in front of the lens. Removing such images first makes sense to me as it’s usually the first step in my culling process when looking through any unedited selection of new photos. Get rid of all the shots which ‘didn’t come out’ and start from there.
They then needed to teach Clips the basics of composition, stability, sharpness and framing. Giving it a concept of the aesthetic value of the positioning of an object, such as a face, within the frame.
When you take a photograph of a person, you often recognise who they are and recognise when they are ready to be photographed by their instinctive smile or a change in pose. Clips is trained to achieve a similar familiarity with faces by remembering those who appear most frequently in your images and whether they appear consistently well-framed as an intentional subject rather than as a casual bystander in the background.
A big part of the editing process for any event photographer is selecting a few great images from hundreds or even thousands of photos taken on location. We instinctively know how to do this, but the AI needed to be taught the value of concepts such as time passing, changes in prominent colours and an idea of how many people are present needing to be photographed.
Autonomous cameras have been tried before with varying levels of success, but even if Clips ends up as nothing more than an interesting detour in the history of photography, the process of teaching it how to take good photos has a lot to teach us about how we, as humans, can create more compelling images of our own.
Source:- https://www.forbes.com/sites/paulmonckton/2018/01/30/google-clips-available-now-trained-by-photographers/#4fc3f9bc2c29
