Google released its Cloud Video Intelligence API to the world today by making it available in public beta, as part of the company’s ongoing push to make AI accessible.
The Video Intelligence API is designed to let users upload a video and get information back about what objects are in it, using a system called label detection. With this release, the company also added support for detecting pornographic content, making it possible to use the service to spot videos that would be inappropriate to share with an audience that isn’t looking for that sort of content.
In addition, Google also announced a number of improvements to its Cloud Vision API to make various features more accurate. The label detection model, which names objects inside an image, now supports more than 10,000 different entities, so it can spot the difference between “breakfast cereal” and just “breakfast.” That model is also twice as good at recall, which means that it’s more likely to pick the most relevant label for an image.
The service’s safe search model, which detects adult content, saw a 30 percent reduction in errors. The Vision API’s text detection model saw a 25 percent increase in average speed of detection, and 5 percent increase in accuracy on latin languages. Google’s system is also better at reading human emotions: the face detection system is more than twice as good at recognizing sadness, surprise and anger than it was at launch.
Google’s services are designed to make it easier for people to implement AI capabilities inside their applications without building the machine learning systems needed to power them. Today’s news shows one of the key benefits of those systems: it’s possible to gain major improvements in applications that use them without doing anything, just because the company behind the system makes improvements in the background.
The Cloud Video Intelligence API launched in private beta earlier this year, as part of the announcements made at the Google Cloud Next conference.
Google is competing with a wide variety of companies in the intelligent API space, including titans like Microsoft, Amazon and IBM.
As part of the Video Intelligence API’s public beta launch, Google announced pricing for the service. Label and adult content detection is free for the first 1,000 minutes of video uploaded, and costs 10 cents per minute for the next 9,000 minutes. Shot detection, which finds scene changes within a video, is also free for the first 1,000 minutes, and then costs 5 cents per minute for the next 9,000 minutes.
Companies that need additional time should contact Google for additional pricing information.