The internet has already made searching for images a breeze. Just go on any search engine, type in what you’re looking for, and voila: Thousands of images will instantly appear before your eyes. But that sort of luxury hasn’t found its way into video, a medium that has seen astonishing consumption rates over the last few years.
Google wants to change that. The Mountain View, California-based company announced a new video intelligence application program interface (API) today at its Cloud Next conference in San Francisco. The tool uses machine learning to extract “entities” (nouns) from videos to make them searchable. It will even mark scene changes. Traditionally, elements related to a specific video had to be tagged manually for them to appear in a search.
The demo at the conference showed the API identifying a dachshund from a short commercial, and figuring out what it was watching was indeed a commercial. In another demo, a search for “beach” loaded a bunch of videos the machine learning software determined had beaches in them.
Google lets you try out the API in a limited demo that shows off its impressive capabilities. Just go on its website and select a video sample. From there, the API will start analyzing the scenes. When its done, it spits out a percentage for the most relevant labels in the entire clip, and the most relevant labels for the parts of the video it determines are separate scenes.
The new API was built to help understand the overall content in videos, and extract actionable insight from that data. Google will hope this new tool will help it catch up to both Amazon and Window’s cloud-based offerings.