Video search becomes Google's next challenge after image search, and is still under development

According to foreign media reports, for Google, identifying objects in pictures is no longer a challenge. Its next challenge is to identify and search the content in videos, because videos have a larger amount of data than pictures and are more difficult to summarize and identify.

According to foreign media reports, identifying objects in pictures is no longer a challenge for Google. Its next challenge is to identify and search the content in videos, because videos have a larger amount of data than pictures and are more difficult to summarize and identify.

Not only Google, but also Social networks Facebook, snap app and Weibo site Twitter are also trying to identify video content.

However, for a long time, YouTube, a video site owned by Google, has been known as the world's second-largest search engine, after Google's search engine. Although text search requests can help Google return YouTube video search results, real video content search technology has not yet been used.

A few months ago, Google presented a gift to the field of scientific research: the YouTube-8M video dataset. Today, Google updated this dataset again. This may not be a coincidence. Open data, especially the Stanford ImageNet database and Microsoft COCO database, is driving research on image recognition search. Artificial intelligence systems need data to become smarter. These organizations are already working to provide such raw data.

Of course, Google does not just want to create advanced technology to benefit all mankind. It also hopes to improve its products with this, just as it introduced Smart Replies to its Gmail mailbox system and added "real-time visual translation" to its Google Translate translation service. Google certainly hopes that YouTube will become the best place for people to search for videos.

"It would be cool if it could recognize videos of cows jumping over the moon or cats jumping over fences." Google senior researcher Jeff Dean said at Google's first TensorFlow Dev Summit.

Dean said this means Google no longer needs to rely on metadata such as descriptions and reviews to conduct searches. This video search technology can also improve video recommendation services.

At this time, we don't know when YouTube will launch this video search feature.

In general, Dean said,"video search may lag behind image search by years."

Editor: Nancy