Abstract

Video thumbnails are often the first thing a viewer sees when browsing or searching for videos. A frame that is visually representative of the video is typically selected and used as a thumbnail representation of the video. Sometimes, such a thumbnail is not an adequate semantic representation of the video. Further, it is possible that such a thumbnail is not visually pleasing. This disclosure describes deep learning techniques to select video thumbnails that are visually attractive and reflect the content of a video. Thumbnails as described in this disclosure are attractive, improve a likelihood of user selection, and help users find relevant content easily.

Share

COinS