I'm sure this is already done/being done. But what the hell.
Today's video search is based (only?) on the text that the uploader describes a video with. Can video search be smarter by recognizing objects in the video? Give me results based not only on the text that the uploader has included but also using the content in the video.
Today's video search is based (only?) on the text that the uploader describes a video with. Can video search be smarter by recognizing objects in the video? Give me results based not only on the text that the uploader has included but also using the content in the video.
Forget video. Even image search (for better search terms, use CBIR (content based image retrieval)) is not possible using content. There are a lot of papers but I doubt how well and fast they work for web-scale data...
ReplyDelete..but even if someone does that for you, how do u propose you search using "content"? upload a pic/video of the object (that means do I always have the pic/video of all the objects that I wanna search with me?) or you type the name of the object and the search recognizes that automatically?
ReplyDeleteBoth of them are tough problems to solve. The second one seems feasible but someone needs to tag the recognized object with a text tag to make it actually visible in search. Or, if you say that the recognized object should automatically be assigned a text tag, that is very very hard problem AFAIK.. may be someone with a better knowledge of computer vision can confirm..