 Welcome back everyone. Today we're talking about the video to faces utility in Surugi Linux. And just like it sounds, it extracts faces from video files. You can find it under applications, Surugi, picture analysis, computer vision, face recognition video, and then video to faces. And it is just a command line tool, just like most of the other tools we've been using. This one's very straightforward. You just simply give it a input file and an output directory. It will take the input file and process it and then try to extract faces from each frame of the video. So you can see here we have the help menu. Video to faces does take a while, so I've already pre-processed everything. If you just do dash H, you get the help menu, which is this. We have the dash I for the input and then dash O for the output. And the output should be a directory. If the directory doesn't exist, the video to faces utility will create the directory for you. And then to run it, it's fairly straightforward. We have video to faces dash I, and then I'm just have a mp4 video file on my desktop. And then I'm outputting also to the desktop into a folder v2f out. Now let's take a quick look at this video playback dot mp4 video. So I have this video here and it's basically just stock. Now the first clip in this stock video is just feet walking around. So we shouldn't find any faces within that video. But if we scroll ahead of here, okay, some more kind of low shots, some high shots, let's say a security camera very far away. This video is probably too far away to actually detect faces, but we can still give it a try. And by the way, this video is in 720p. So not totally high def, it would be a kind of a lower end security camera resolution. And here we have people walking, but it's basically the back of their heads, not their faces. More people kind of side. And side walking again. But it's a little bit fuzzy. And then we have some phases here that we might be able to detect. And then we get a couple more front shots like this has a lot of people with front shots, some of them are blurry, some of them are more clear. So we have a couple of different problems here, people walking around, we might be able to detect some of the profile images in that. So the idea behind this is we go through frame by frame, and then just do face detection on each individual frame. So the idea is if you have a video that has a lot of space with no people in it, and then somebody like walks into the room, then you should be able to just extract their facial features from those particular frames. And then the rest of the time, there's nothing to extract. So this could be really interesting for especially security camera footage where you're looking at a door or a garage for a very long period of time. And then you want to know when someone came in and who they were and extract their image. So whenever we run this, it's just video to faces and then the input file, so dash i, the file that we have dash o, the file that we're outputting to, and then it starts to just extract each frame and then say how many faces it found in the picture. Remember, the first part of my video was just feet walking around. So no faces in that. Now, whenever it does detect something that a face was located at pixel location top, and then it gives the location in the particular frame, unfortunately, it doesn't say which frame it was found in. But basically, a face was detected, and it'll say how many faces it detected. I can say that these two were false positives just happened to match the pattern. And I'll show you those in a second. And then we get down later. Remember, we had a bunch that didn't have anything. And then we start to see that a lot of faces are detected. And that's where the scene where people were walking straight at the camera. And we actually had a good clear view of their faces, it was able to find multiple faces in the picture at the same time. So let's go ahead and see what this output looks like. You get a directory that's just full of PNG images. And if you double click on them, these first two, like I said, were probably false positives, just a random thing that just happened to match the pattern of the model that we're using here, three were false positives. And then we start getting into a real face. So this is a blurry image of somebody's face that was picked up, we get some better images down below. Basically, all I had to do was just feed the video to this program. And then it starts to extract all of these these faces, then we can use, for example, facial recognition, or just use the image to try to identify people if we wanted to do that. So it's a very easy, quick way to extract faces from whatever video it is that you have. So I want to show you another situation. First, I had a 720p video, and that's what we analyzed first. This next video is high depth, pretty interesting. It's just a camera, but you can see that there's a lot of very clear faces in the video. However, we have some statues, and then we have some people in the background, we have a lot of different shadows and things going on here. But basically, people walk in a very predictable way on both sides. This video is very high depth, and it took a very long time to process. Now let's see what the results of that were. When we process that video with video to faces, it did find a face very consistently in the entire video. And I let this run actually for a long time in another test. And basically, I kept getting the same results. So let's go see what those results actually were. This is pretty much what I got the entire time. Now it did detect a face. However, it was the face in the statue at the very center of the video. It really consistently found that statue in the center there, but it didn't find any of the faces of the people walking by. So what we need to do is think about what our video looks like. This camera is set there all the time. And if I wanted to try to do analysis on that particular camera, I'm going to have to mask away or I could mask away those statues and focus my analysis just on the area where the people are walking, for example. Or if I'm not analyzing the camera stream directly, then I could just take a clip and then edit that clip and crop it down to just the areas that I'm interested in analyzing. Video to faces in the Surugi Linux distribution is an interesting, quick, easy to use tool. And I can think of a lot of cases that it would really benefit. I hope that was useful. Thank you very much.