Hello and welcome back. Let us now talk about Video Inpainting. You basically already have all the tools to do Video Inpainting. Remember, and I'm showing these in very slow motion, a video is a sequence of frames, a sequence of images. So, we can basically use the tools that we developed for Image Inpainting now in Video Inpainting. How? For example, we can take, in the Cut and Paste, instead of two-dimensional patches x, y, we can consider three-dimensional patches x, y, t and look at them in the video. In the whole video, that would be very expensive but we can look at them in segments of the video. We can also run the partial differential equations or the variational formulations considering, for example, three-dimensional gradients. When we talk about curvature, we can consider the videos as surfaces and talk about surface curvatures like Mean curvature or Gaussian curvature. So basically, the same type of concepts, Laplacian, we can take derivatives in the x, y, and t directions. Some type of concepts, we can expand to do Video Inpainting. And that's how today, Video Inpainting is done in the literature extending those concepts that we saw for images. Now, Video Inpainting has a bit of different characteristics. For example, here, let me just show you with the arrows. There are some objects that are static in the video so they will never appear in any of the frames. And there are some objects that move in the video so it's slightly different if we want to remove static objects than if we want to remove moving objects. For example, if we want to remove a moving object, all what we have to do is wait until it's gone in one of the future frames and copy from those frames into the current frame. So, I want to remove this now. Look at my pointer, now it's gone. So, I basically can copy those pixels, those pixels from future frames into the current frame. Because the object is moving, the current background appears in some of the previous or some of the future frames. When the object is fixed, it's slightly different. The background is never revealed but this Cut and Paste or partial differential equations or variation of formulations will work there. So, you could if you want make a distinction between these two and differently inpaint each region or you could just let those techniques that we described for images work with all the objects. Here, we see an example of this video inpainted with those combination of Cut and Paste propagation of information as we have seen, we basically have removed a number of objects, we have removed basically this, this, and this. We have removed both static and dynamic objects with those techniques without making any explicit distinction between them. Here is another example. what we want to remove now, as you see, is this object. Remember, for inpainting you get the video, as we got the image, you have to get the mask, a video with the mask. Now, of course, doing that by hand for video is much more time-consuming than doing it by hand for still images. But we have to mix to do segmentation in video as we have seen, for example, the Roto Brush in Adobe's After Effects. And here is the, the result of basically removing this person or this mask. And you see the video looks completely natural. We, it's very hard to realize for us that this is not the original video but it actually is not and it's actually inpainted with this dynamic object. And now here, we have a second example. This is an example that we have already seen in the past and once again, we are basically removing one object and making you believe that it was the original video. This basically concludes our presentation of Image and Video Inpainting. We have seen very, very good techniques for doing this and as we talked in the past, this is used a lot, for example, in the movie industry. But its also used a lot to improve our own pictures, our own images, and remove objects that we don't like on those images. I hope you had fun, this is really a fun topic. You can make a lot of nice stuff with Image and Video Inpainting. Thank you very much and I am going to see you again next week. Thank you.
