What are you watching and why do encoders care?
Understanding video viewing habits is big business for Hollywood and content providers. It not only helps them to tailor their websites for accessibility and to improve the user experience, but also to cash out on consumers’ viewing time. In the article below, we see that even Facebook is getting its feet wet with a recent test for video advertising. As the article notes, “If the test is judged to be successful, Facebook will be in a good position to capture a share of the lucrative U.S. television ad space — which is expected to hit $81.6 billion in 2017, according to PwC.” Read the full article here.
The introduction of video ads could enable Facebook to reinvigorate its advertising revenue. The article also notes that Facebook is being careful not to introduce so many video ads that users get annoyed and leave the social media mogul. One would imagine that, in addition to the sheer number of videos in users’ news feeds, another factor affecting users’ acceptance of the videos ads would be the quality and accessibility of the videos themselves. As many devices now are capable of displaying high quality video, consumers will be expecting both ease of accessing videos and a good viewing experience.
Working in the video compression industry, we at Euclid Discoveries are often asked how we compress a few “standard test videos” (abbreviated here as STVs) that are known to be notoriously difficult to compress well. Over the years, experts in video compression have promoted these STVs because they stress the limits of encoders and effectively differentiate encoder performance. During encoder development, it is hard not to get caught up in how one’s encoder performs on these STVs. However, one must take a step back and ask: are these the types of videos that people watch? Are the STVs representative of the types of videos Facebook will start inserting into users’ news feeds?
We believe the answer is mostly “no.” The STV set explores the limits of encoder performance in many different facets. Measuring encoder performance on the full STV set provides a comprehensive evaluation of the encoder’s strengths and weaknesses – from an engineering point of view. However, we would argue that encoder performance should be weighted toward the type of videos that people typically watch, not equally weighted across many types of (unequally important) videos. As an example, several videos in the STV set show slow panning across nature scenes of high complexity. These types of videos are difficult to compress well, for reasons that will be explained in a future blog post. Yet, we must ask: how often do people watch that type of video? At best, one can imagine such a clip as a transition shot in a movie, representing a tiny fraction of the movie’s video content.
What then are the types of videos that people do watch? Music, sports, gaming, and movies comprise four of the top five video categories on YouTube, and they all contain one common characteristic: motion. Just as we live in a world full of movement or motion, consumers are drawn to videos with motion as well. While music has been around since the Stone Age, music videos typically incorporate flashy, rhythmic, or distinctive motions to capture and then retain the attention of modern viewers. It is also no accident that in the video age, visually appealing, high-motion “TV sports” such as football and basketball have overtaken “slower” sports such as baseball in popularity. And a quick perusal of the top grossing movies of 2013 shows that eight of the top ten movies can be categorized as action/adventure, again full of motion. It is no accident that the video Facebook selected for its video ad test is a movie trailer, usually a condensed version of the high-action (high-motion) scenes in a movie.As it turns out, motion is another differentiator of encoder performance, as high-motion videos are also difficult to compress well.
A portion of the STV set of “difficult” videos does contain high motion, but we believe that any test set that truly reflects the viewing habits of today’s generation would contain 80% or more high-motion videos. At Euclid Discoveries, we are focused on improving encoder performance for high-motion videos in particular, the type of videos that people are watching. In a future blog post, we will detail the challenges of motion for video compression, as compared to other types of challenges, and possible approaches to handling the challenges of high-motion video.
In the meantime, we wish you happy viewing of our world in motion in 2014!