Click on any phrase to play the video from that point.
[♪ techno music ♪] [Adobe TV Presents...] [♪ techno music ♪] [Prototype] Hi, my name is Wil Li, and I'm a researcher in the Creative Technologies Lab at Adobe, and today I wanted to demo some technology for creating seamless transitions in interview footage. Now one of the most common types of video that people work with these days is interview footage and that's a video of somebody talking, like me answering questions or maybe describing some idea, and we see this type of footage a lot in things like news stories or documentary films, and in a typical scenario what happens is that you capture a lot of interview footage and only end up using very little of it. We want to use maybe 1 or 2 minutes for a particular piece, and so in order to go from that large amount of footage down to 1 or 2 minutes, we end up cutting out most of the video, and and also maybe removing certain words that we don't want in there, maybe splicing 2 sentences together that should go together. Now the problem is whenever we create, or whenever we make these kinds of edits, we introduce really noticeable jump cuts that can be quite distracting, and so the technology that we've developed is useful for helping bridge the gap across these jump cuts, to make them not noticeable. So the way that we do that is we take footage that would usually end up on the cutting room floor, and we try to take the necessary frames in order to bridge the gap from the beginning of the cut to the end of the cut so that the transition looks seamless. So now let's take a look at some of the actual examples that we've used and we've run using our system. Okay, so on the left here, what I'm going to do is I'm going play a clip from a video, and you'll notice a couple of jump cuts here. So we've spliced out--we've cut out some words here that the person's saying, and we've maybe removed a pause, so let me just play this video, and I'll play it a couple times so you can see where the jump cuts are. So on the left here, we'll see this person talking. Right there, there was 1 jump cut, and in just a couple of seconds you'll see a second one near the end of this clip. [silent video playing of man talking] There was the second one. So those 2 jump cuts are really noticeable, and we'd like as much as possible to get rid of them. Now if we look on the right, we can see the result of running our algorithm on this clip. [silent video playing of man talking] So this is the same clip, and instead of seeing these really noticeable jump cuts, what we get is what appears to be a fairly seamless single take of this section of the video. [silent video playing of man talking] And, hopefully, if I hadn't told you ahead of time that there were 2 jump cuts in here, you wouldn't notice at all that there was anything strange going on. This would look like just a single take. All right, let's look at a couple of other examples as well. So here at this time I'm actually going to do something different. I'm going to play our result first, and, hopefully, what this is going to look like is a continuous take, and then I'll go back and show the input, and we can see where the jump cuts actually were. [silent video playing of man talking] So, again, this is footage that we took from a real interview. [silent video playing of man talking] In this case, we removed some pauses and some unnecessary words in order to create this little clip. [silent video of man talking] Okay, now let me show you the input, and here, hopefully, you'll notice that there are actually several jump cuts in this sequence. There was 1. [silent video playing of man talking] Right there, there was another one and another one. I think we just removed an "um" in that case, and there's one more right there. So in this case, there were actually 4 different jump cuts here, and let me just play the result 1 more time so you can see what that looks like. [silent video playing of man talking] So as you can see the really noticeable kind of sharp movements of the head are no longer there, and for the most part this looks like a seamless take. [silent video playing of man talking] Okay, and let me finish up with 1 more, actually more challenging example. So you'll notice that in this video, one of the challenging things is that the person uses his hands quite a bit, and it's quite--it can be quite challenging to try to bridge these cuts across you know where there are lots of hand motions, but you'll see that our algorithm manages to do a pretty reasonable job. [silent video playing of man talking] So here you can see he's gesturing. He's using his hands, using both hands. He's also pretty expressive I would say. His facial expression changes quite a bit, maybe compared to the last 2 examples. [silent video of man talking] Okay, so that was the result, and let's go back and look at the actual input to see where those jump cuts originally were. [silent video of man talking] So there that first one was pretty obvious. His hands are there and all of a sudden they disappear. [silent video playing of man talking] In this case, I think-- he's does another kind of funny motion with his hands. Yeah, there was another jump cut. Yeah, so in this case, there were also 3 jump cuts in the original input video that we managed to create seamless transitions for. Okay, so that's the technology that I wanted to demo today. We think it could be really useful for video editors who are working with interview footage, especially since hiding jump cuts is one of the most time consuming and annoying tasks that editors have to perform. We'd love to integrate this technology into some of our existing products to get it into the hands of our users. Thank you. [♪ techno music ♪] [The seamless edits technology is based on a collaborative research project with the] [following contributors: Floraine Berthouzoz (UC Berkeley),] [Maneesh Agrawala (UC Berkeley) Wilmot Li (Adobe Systems)] [Executive Producer Bob Donlon, Producer Karl Miller, DP/Post Erik Espera] [Director/Motion Graphics Kush Amerasinghe, Adobe TV Productions, tv.adobe.com]
