Would you like to inspect the original subtitles? These are the user uploaded subtitles that are being translated: 1 00:00:00,420 --> 00:00:03,090 Hello and welcome to this new tutorial. 2 00:00:03,090 --> 00:00:03,410 All right. 3 00:00:03,420 --> 00:00:07,410 So in the previous Statoil we created our as is the neural network. 4 00:00:07,470 --> 00:00:10,800 So now we have the frame and the net the neural network. 5 00:00:10,830 --> 00:00:14,960 So we have one less thing to create before we are ready to apply the detect function. 6 00:00:15,000 --> 00:00:16,790 It's the transformation. 7 00:00:16,850 --> 00:00:19,180 So that's exactly what we're going to do in the Statoil. 8 00:00:19,270 --> 00:00:21,350 We're going to create that transformation. 9 00:00:21,350 --> 00:00:26,710 I'm saying create because we're actually going to create a new object of the base transform class. 10 00:00:26,880 --> 00:00:32,820 So it is exactly this object that will do the transformation itself on the image so that this image 11 00:00:32,910 --> 00:00:38,520 is compatible with the neural network that is this transformation will make sure that this frame can 12 00:00:38,520 --> 00:00:41,080 get in to the neural network net. 13 00:00:41,100 --> 00:00:41,430 All right. 14 00:00:41,440 --> 00:00:42,300 So let's do this. 15 00:00:42,300 --> 00:00:44,450 It's going to be very easy and fast. 16 00:00:44,490 --> 00:00:47,360 We just need one line of code because we have the base friends from class. 17 00:00:47,370 --> 00:00:51,120 We just need to create an object of this class so let's do this. 18 00:00:51,120 --> 00:00:55,610 We're going to call this transformation transform obviously. 19 00:00:55,770 --> 00:01:03,160 And since this transform transformation is going to be an object of the base transform class. 20 00:01:03,180 --> 00:01:07,100 Well I'm calling this class and now we have to input several arguments. 21 00:01:07,170 --> 00:01:14,510 So the first argument is not that size and not that size is the target size of the images to feed to 22 00:01:14,520 --> 00:01:15,700 the neural network. 23 00:01:16,020 --> 00:01:24,930 So here we go net that size that the second argument is a couple of three arguments a triplet is going 24 00:01:24,930 --> 00:01:31,520 to be a triplet of three numbers that will allow to put the color values at the right scale. 25 00:01:31,830 --> 00:01:33,320 And what is this right scale. 26 00:01:33,360 --> 00:01:37,550 Well that's exactly the scale under which the neural network was trained. 27 00:01:37,620 --> 00:01:43,770 That is this new will network from which we're losing the weight was trained it was trained under some 28 00:01:43,770 --> 00:01:47,540 certain convention and part of this convention concerns the skill. 29 00:01:47,790 --> 00:01:53,680 And what we're doing now is exactly putting the right scale for the color values. 30 00:01:53,700 --> 00:01:58,940 So now I'm just going to put three numbers don't worry about the numbers. 31 00:01:58,950 --> 00:02:01,100 These are just numbers to get the red scale. 32 00:02:01,230 --> 00:02:02,820 But it's not the most important. 33 00:02:02,850 --> 00:02:10,890 So these numbers are the first one is 104 divided by 256 point zero. 34 00:02:10,890 --> 00:02:23,220 Then the second number is 117 divided by 256 point zero and the third and final number is 123 divided 35 00:02:23,220 --> 00:02:26,550 by 256 point zero. 36 00:02:26,850 --> 00:02:27,450 All right. 37 00:02:27,450 --> 00:02:31,910 So net size is the target size of the images to be given to the neural network. 38 00:02:32,100 --> 00:02:37,980 And these three values here are some scale values to make sure that the color values are in the right 39 00:02:37,980 --> 00:02:40,200 scale and that's it. 40 00:02:40,380 --> 00:02:42,900 Actually our transformation is ready. 41 00:02:42,900 --> 00:02:50,280 So now time for some exciting stuff in the next tutorial we will actually open the video then we will 42 00:02:50,310 --> 00:02:58,440 iterate on the frames of this video because I remind that this technique is a frame by frame detection. 43 00:02:58,500 --> 00:03:05,130 So we playing the detect function on each frame of the video you're going to see that this two seconds 44 00:03:05,130 --> 00:03:10,840 video has sixty eight frames I think something like that 67 or 68. 45 00:03:10,980 --> 00:03:16,340 And we're going to play the detect function on the 68 frames of this video. 46 00:03:16,560 --> 00:03:21,360 So the first thing we'll do after opening the video is that we'll get all these frames. 47 00:03:21,510 --> 00:03:24,210 Then we'll apply that to check function on each of these frames. 48 00:03:24,360 --> 00:03:30,510 Then there's the deck function will detect some dogs humans or whatever on the frames will print the 49 00:03:30,510 --> 00:03:36,370 rectangles on each of these frames and then we will reassemble the whole frames to make a new video. 50 00:03:36,450 --> 00:03:41,180 That is the original video with the detector rectangles detecting the objects. 51 00:03:41,370 --> 00:03:43,920 So I can't wait to do that in the next tutorial. 52 00:03:43,920 --> 00:03:45,810 We're about to see the final video. 53 00:03:45,810 --> 00:03:47,280 Can't wait to show you this. 54 00:03:47,340 --> 00:03:49,200 Until then enjoy computer vision. 5714