One of the biggest things that holds back many new photographers, or creators in general, is the fear of being harshly judged by the public.
Anyone who has created something knows this one salient fact to be true: The world is filled with harsh critics.
Sometimes, there’s something to the criticism, however, and it’s rarely wise to ignore these critics.
Google’s Neural Image Assessment AI hopes to soon join the ranks of the “helpful” set of judges as the advanced technology can now rank images based upon the picture’s aesthetic and technical qualities.
This is not Google’s first attempt at using AI to evaluate photos: The Silicon Valley giant has previously deployed convolutional neural networks, or CNN, to judge images within specific categories such as landscapes.
These evaluations were binary in nature when compared with NIMA’s new approach. The latest method allows for the evaluation of photos across a range of categories applicable to all photos regardless of type.
NIMA uses a CNN that is trained to estimate what photos a user would find aesthetically pleasing or technically proficient on a scale of 1 to 10. Coming up with this number relies upon object recognition networks which help the AI recognize elements within the image.
The system can catalog and grade images at a rate much more efficiently and reliably than a human. This technology could see potential use in automated image editing and sorting, for example. Other applications down the road include real-time photo editing or suggestions for other optimization aspects.
To evaluate NIMA’s accuracy, the team compared the AI-assigned score with a human-assigned score to see the deviation between the two. The close proximity of the mean scores bodes well for the AI’s ability to recognize images in a similar way as humans when judging them on a technical or aesthetic aspect. Further, the artificial intelligence is still a young science, meaning potential future innovations could lead to even better functioning.
I’d rather be harshly judged by the public than by a computer! A computer can tell me a horizon is crooked or that I have a color cast, but it can’t “judge photography”.