Artificial intelligence that can understand what the video is about.
We caught a glimpse into the future of online search, and this is what we saw there: a 17 second video with a puppy provided by Clarifai, a small startup specializing in artificial intelligence.
The video shows a cute puppy, poking at his mistress, but the most interesting is in the lines below the video. I use a database of 10,000 visual categories that Clarifai collected over the past half a year, the company's software monitors every frame of the video, automatically describing them with words such as dog, woman, eyes and even cute.
')
The basic idea is that you can then search for these words and the software will show those frames where images appear corresponding to them.
This part of the trend in AI, called “deep learning”, widely used by technology giants, allows you to achieve human recognition levels. Google uses deep learning to improve voice recognition in Android. Microsoft uses it to instantly translate Star Track language from language to language. Facebook with its help improves the automatic recognition of people in the photo. And, soon, deep learning will change how we search for video, making it possible to automatically analyze clips with a recognition of what is happening to them.
This is very important because, at present, video search technologies produce results based on text that is contained outside the video - headlines, comments, or metadata, or using tags attached to the download. With the help of these technologies, you can find a video of a car chase, but to find a moment when it starts is not possible.
Clarifai's director, Matthew Zeiler, says they can solve this problem. We wrote about Zeiler last year, when its AI algorithms won the prestigious ImageNet image recognition contest. Currently, Clarifai is already selling high-quality image search software that can be used on smartphones, companies providing catalogs or anyone who works with large amounts of images.
Video search technologies can bring in deep learning for use in video archives or even in companies like GoPro. Imagine that you can get all the moments when you are off the ground from a video taken during a ski holiday. Or search for a garden gnome robber from your lawn on video from surveillance cameras. We will be surprised if, in the course of the coming years, you will not be able to do a similar search, including via YouTube videos.
By that time, Clarifai software will most likely be able to automatically process and generalize videos and even sort them according to what happens in them. “This is just the tip of the iceberg,” says Zeiler.
Best wishes generate.club
The video shows a cute puppy, poking at his mistress, but the most interesting is in the lines below the video. I use a database of 10,000 visual categories that Clarifai collected over the past half a year, the company's software monitors every frame of the video, automatically describing them with words such as dog, woman, eyes and even cute.
')
The basic idea is that you can then search for these words and the software will show those frames where images appear corresponding to them.
This part of the trend in AI, called “deep learning”, widely used by technology giants, allows you to achieve human recognition levels. Google uses deep learning to improve voice recognition in Android. Microsoft uses it to instantly translate Star Track language from language to language. Facebook with its help improves the automatic recognition of people in the photo. And, soon, deep learning will change how we search for video, making it possible to automatically analyze clips with a recognition of what is happening to them.
This is very important because, at present, video search technologies produce results based on text that is contained outside the video - headlines, comments, or metadata, or using tags attached to the download. With the help of these technologies, you can find a video of a car chase, but to find a moment when it starts is not possible.
Clarifai's director, Matthew Zeiler, says they can solve this problem. We wrote about Zeiler last year, when its AI algorithms won the prestigious ImageNet image recognition contest. Currently, Clarifai is already selling high-quality image search software that can be used on smartphones, companies providing catalogs or anyone who works with large amounts of images.
Video search technologies can bring in deep learning for use in video archives or even in companies like GoPro. Imagine that you can get all the moments when you are off the ground from a video taken during a ski holiday. Or search for a garden gnome robber from your lawn on video from surveillance cameras. We will be surprised if, in the course of the coming years, you will not be able to do a similar search, including via YouTube videos.
By that time, Clarifai software will most likely be able to automatically process and generalize videos and even sort them according to what happens in them. “This is just the tip of the iceberg,” says Zeiler.
Best wishes generate.club
Source: https://habr.com/ru/post/376483/
All Articles