Organizing a photo library containing thousands of RAW files, photos, and videos can be a time-consuming and mind-numbing task. The mere thought of manually tagging, grouping, and organizing photos and videos can give even the most patient photographers shivers. Good thing then that we live in the age of artificial intelligence (AI) that can take the burden of boring tasks off our shoulders. For anyone who is not keen on spending countless evenings fighting a losing battle of keeping their photo libraries neatly organized, PhotoPrism [1] might be manna from heaven. This application pairs a polished yet capable user interface with the powerful TensorFlow machine learning engine, providing you with a platform that can take care of keeping tabs on your photos and videos as if by magic. That description is not an exaggeration: The inner workings of TensorFlow are hidden away from the user's sight so well that it's almost too easy to forget that PhotoPrism is powered by a complex piece of software engineering and not fairy dust. Of course, TensorFlow is not perfect, and sometimes it produces some hilarious results. But they are pretty much the only telltales of what actually powers PhotoPrism.

TensorFlow and PhotoPrism

PhotoPrism relies on TensorFlow to perform three important tasks. The first task is image classification. To simplify, TensorFlow analyzes images and assigns relevant labels to them. For example, all architectural photos get the Building label, and wildlife photos may get various labels, depending on the main subject (for example, Bird, Butterfly, etc.). The second task is face recognition, which allows PhotoPrism to find similar faces and group them into clusters. The third task is detecting images that are usually referred to as "not safe for work," which in most cases means photos containing nudity.

Deploying PhotoPrism

Because PhotoPrism consists of several complex moving parts, deploying it would be a rather daunting proposition if it weren't for containers. PhotoPrism is distributed as a Docker container image that reduces the task of getting the application up and running to a few simple steps.

[...]