1 00:00:01,320 --> 00:00:03,480 Right to touch with our own image. 2 00:00:03,480 --> 00:00:07,010 We can go online and download some images. 3 00:00:07,050 --> 00:00:09,240 I'm just gonna download some cuts images 4 00:00:14,720 --> 00:00:19,040 I'm gonna go to Google dot com my computer. 5 00:00:19,440 --> 00:00:24,400 I don't know it's running a bit slow. 6 00:00:41,580 --> 00:00:42,110 Okay. 7 00:00:42,210 --> 00:00:43,320 We can download a few 8 00:00:46,920 --> 00:00:49,470 cannot right click this save this 9 00:00:53,050 --> 00:00:54,790 occur this empty one 10 00:00:58,840 --> 00:00:59,480 then. 11 00:00:59,550 --> 00:01:00,540 How about this one. 12 00:01:00,570 --> 00:01:05,560 Save this I empty too. 13 00:01:06,690 --> 00:01:07,750 Let's try this. 14 00:01:07,760 --> 00:01:11,590 Save this I empty three. 15 00:01:11,800 --> 00:01:13,780 How well we probably. 16 00:01:13,860 --> 00:01:14,360 Mix it up. 17 00:01:14,370 --> 00:01:17,160 I'm gonna search car image as well. 18 00:01:20,320 --> 00:01:27,290 NOBODY'S SAFE THIS I empty for a c 19 00:01:33,530 --> 00:01:34,180 c of us. 20 00:01:34,940 --> 00:01:38,580 I am G five K. 21 00:01:39,830 --> 00:01:41,140 Oh we try a bird. 22 00:01:41,230 --> 00:01:43,600 We got lost to image 23 00:01:54,830 --> 00:01:56,350 I empty six. 24 00:01:56,410 --> 00:02:01,720 Okay so let's take these images and put them in our images folder. 25 00:02:01,930 --> 00:02:02,980 Come here. 26 00:02:02,980 --> 00:02:08,520 Fetch these. 27 00:02:08,870 --> 00:02:10,640 This is my project forward. 28 00:02:10,690 --> 00:02:16,120 I've got an empty photo here called images right. 29 00:02:16,790 --> 00:02:24,780 So let's test the images and then I can close this. 30 00:02:25,370 --> 00:02:32,750 Okay so I'm gonna come down here and um I'm gonna see if this actually to create a new safe point. 31 00:02:32,840 --> 00:02:34,040 Call this number eight. 32 00:02:39,360 --> 00:02:49,530 And I'm gonna see um my image a cause and we can just take one of the images. 33 00:02:49,530 --> 00:02:56,400 Let's start with S. I MDG 1 the JPEG that S J peachy like this. 34 00:02:56,580 --> 00:03:08,220 And then the name of the file we know we can locate it in our images folder in the name of the image 35 00:03:08,400 --> 00:03:13,590 is my image name right. 36 00:03:13,630 --> 00:03:21,440 So once that is done we're gonna see it image a course and Pitot array and then we're gonna do. 37 00:03:21,480 --> 00:03:31,680 And the image that I am read then we pass the file name and the path and then flatten 38 00:03:35,110 --> 00:03:36,040 it course force 39 00:03:39,850 --> 00:03:43,960 once that is done we gonna see my image. 40 00:03:47,230 --> 00:03:55,280 Because Sy pi dot miscellaneous dot image resize 41 00:03:58,430 --> 00:03:59,740 we need to resize it first. 42 00:04:00,590 --> 00:04:12,320 What we're gonna do is post image here and then the size here is gonna be um the size of our standard 43 00:04:12,320 --> 00:04:18,860 image which is this the height and weight of the of the images that we use we're gonna use resize it 44 00:04:18,860 --> 00:04:22,700 to the size num underscore p x number of pixels. 45 00:04:23,240 --> 00:04:26,430 So I'm gonna pass this size over here. 46 00:04:28,570 --> 00:04:32,540 It's gonna be a number of pixels by a number of pixels and then 47 00:04:35,610 --> 00:04:38,490 I'm gonna see a dot reshape 48 00:04:42,830 --> 00:04:53,970 then I'm gonna pass one over here then the number of pixels times number of pixels transpose right. 49 00:04:55,520 --> 00:05:02,630 So this will give us the um the vector that we wanted a column vector which our neural network accepts. 50 00:05:02,720 --> 00:05:06,590 So I'm gonna say my um 51 00:05:10,730 --> 00:05:13,680 C AMG prediction. 52 00:05:14,980 --> 00:05:22,470 Because then I'll call it a predictive function and what I'm gonna do is take the weight by accessing 53 00:05:22,470 --> 00:05:31,340 our dictionary and passing w here to get a weight and then the bias as well. 54 00:05:31,500 --> 00:05:41,260 We find it in the same dictionary b over here and then the input is my image. 55 00:05:41,420 --> 00:05:47,400 Here we can show the predicted image we can show the um the image by saying BLT. 56 00:05:47,430 --> 00:05:49,490 I m show. 57 00:05:50,030 --> 00:05:59,910 And then um we simply pass image we post this here and then we can print the result as well and um we 58 00:05:59,910 --> 00:06:06,740 can simply do print. 59 00:06:07,590 --> 00:06:14,070 The prediction is an SDR. 60 00:06:14,300 --> 00:06:21,300 I'm gonna squeeze the dimension by doing MP dot squeeze and I'll show you later on the difference between 61 00:06:21,300 --> 00:06:23,430 using squeeze and not using a squeeze. 62 00:06:23,430 --> 00:06:32,460 After we run this experiment and what I'm gonna pass us argument here is my empty prediction here. 63 00:06:32,490 --> 00:06:33,210 Right. 64 00:06:33,650 --> 00:06:37,310 So yeah this should give us the resort. 65 00:06:37,560 --> 00:06:39,770 I'm gonna do POTUS show over here 66 00:06:43,970 --> 00:06:51,130 okay so I've passed I am g 1 this is I am g 1. 67 00:06:51,210 --> 00:06:54,290 Let's see what a neural network is gonna say. 68 00:06:54,510 --> 00:06:56,480 Gonna run over here run module 69 00:07:05,510 --> 00:07:06,080 okay. 70 00:07:06,140 --> 00:07:06,910 Creasy. 71 00:07:07,310 --> 00:07:09,520 Test like reach the 70. 72 00:07:09,680 --> 00:07:10,310 Close this 73 00:07:13,610 --> 00:07:14,700 I am Reed. 74 00:07:14,780 --> 00:07:15,940 We have an error. 75 00:07:15,950 --> 00:07:18,410 It says a.. 76 00:07:18,440 --> 00:07:22,220 Image has no attribute. 77 00:07:22,220 --> 00:07:23,150 I m Reed. 78 00:07:23,900 --> 00:07:24,460 Okay. 79 00:07:24,490 --> 00:07:27,530 A C right. 80 00:07:27,620 --> 00:07:29,010 Um yeah. 81 00:07:29,090 --> 00:07:30,350 Let's inspect our code. 82 00:07:31,640 --> 00:07:33,500 Um what we did 83 00:07:37,070 --> 00:07:39,020 I read okay. 84 00:07:39,070 --> 00:07:44,180 We probably have a different version of PSI pi to have to reinstall the. 85 00:07:44,250 --> 00:07:48,340 So the decide package and then um. 86 00:07:48,370 --> 00:07:48,940 Oh [REMOVED] yeah. 87 00:07:48,940 --> 00:07:50,700 Also there's a mistake. 88 00:07:51,070 --> 00:07:56,430 Remember our image has three channels so we've gonna multiply by three over here. 89 00:07:56,830 --> 00:08:06,740 When we reshape it right we have to reshape it to the the column vector that our neural network accepts. 90 00:08:06,970 --> 00:08:07,360 Right. 91 00:08:07,420 --> 00:08:09,700 And we saw how to do this earlier. 92 00:08:10,690 --> 00:08:13,050 So we simply use in this hyper library to reshape it. 93 00:08:13,060 --> 00:08:16,240 But it seems this a library. 94 00:08:16,440 --> 00:08:19,270 No it's not it's not playing ball. 95 00:08:19,330 --> 00:08:19,630 Okay. 96 00:08:19,660 --> 00:08:20,680 Let's see whether it works. 97 00:08:20,680 --> 00:08:21,750 After fixing this. 98 00:08:21,790 --> 00:08:22,420 If it doesn't. 99 00:08:22,430 --> 00:08:29,760 Would have to revert to a different version of spy or click here to run the module. 100 00:08:37,650 --> 00:08:38,100 Okay. 101 00:08:42,910 --> 00:08:46,060 What does it say. 102 00:08:46,060 --> 00:08:47,770 Wait we have something near 103 00:08:54,940 --> 00:08:57,490 it says I am resize is deprecated. 104 00:08:58,060 --> 00:08:58,500 Okay. 105 00:08:58,570 --> 00:08:59,790 Wait. 106 00:08:59,800 --> 00:09:03,400 If you you're still in work then this is what we have to do. 107 00:09:03,510 --> 00:09:06,080 I'll show you perhaps some of you still have the arrow. 108 00:09:06,130 --> 00:09:11,440 You come to the installation folder where we install the python installer here you go to the script 109 00:09:11,600 --> 00:09:19,420 photo within the installation folder and then I'm going to hold on shift right click and then open come 110 00:09:19,420 --> 00:09:25,060 on window here and I'm going to on install side by or do Pip 111 00:09:28,360 --> 00:09:29,440 on install side by 112 00:09:32,310 --> 00:09:34,560 and then you would ask Do I want to continue. 113 00:09:34,610 --> 00:09:38,350 Say yes and then it says successfully on installed. 114 00:09:38,460 --> 00:09:45,910 Then I'm going to install Sai pi one point one cause those image root functions are from Sai Pi I think 115 00:09:45,910 --> 00:09:47,190 get deprecated now. 116 00:09:47,190 --> 00:09:51,480 That is why new installations might receive such an arrow. 117 00:09:51,480 --> 00:09:57,770 Later on we would um we would have to use the other library such as the p. 118 00:09:58,040 --> 00:10:05,760 Uh the python imaging library pill and perhaps not plot lib to reshape and read out what images we would 119 00:10:05,760 --> 00:10:06,530 have to do that later. 120 00:10:06,540 --> 00:10:08,290 But now we can use IPA. 121 00:10:08,490 --> 00:10:13,270 So I'm gonna do pip install pi. 122 00:10:13,830 --> 00:10:19,770 You cause one point one point zero like this then I'm going to hit enter. 123 00:10:21,660 --> 00:10:25,880 So he's gonna take a while to install the um the side pi. 124 00:10:26,310 --> 00:10:34,590 So if you receive an arrow about with regards to um image read then you would have to. 125 00:10:35,030 --> 00:10:39,030 You would have to reinstall CI pi it successfully installed. 126 00:10:39,220 --> 00:10:40,680 You just might take a longer time. 127 00:10:40,680 --> 00:10:41,240 It's fine. 128 00:10:41,760 --> 00:10:42,330 Okay. 129 00:10:42,330 --> 00:10:46,290 So once that is done verify that you have this here. 130 00:10:46,830 --> 00:10:48,060 I'm gonna click here to run 131 00:10:52,430 --> 00:10:53,080 it's learning 132 00:10:57,350 --> 00:10:57,720 Okay 133 00:11:00,660 --> 00:11:01,570 it's what we have. 134 00:11:02,110 --> 00:11:06,960 And this is somewhat image image number one and the prediction is zero 135 00:11:11,460 --> 00:11:12,600 this is strange. 136 00:11:12,600 --> 00:11:16,720 So it failed to predict this accurately. 137 00:11:16,890 --> 00:11:24,000 We have some warning here it says image resize is deprecated and will be removed at one point too. 138 00:11:24,000 --> 00:11:29,910 Okay so we can install we can use SDK major so we can use ASCII image for this next time. 139 00:11:30,300 --> 00:11:34,690 Okay but let's try our neural network once more. 140 00:11:35,220 --> 00:11:43,180 So the first prediction s supposed to be a cut but we the the resort we got was results for no cut. 141 00:11:43,260 --> 00:11:44,020 Let's try. 142 00:11:44,130 --> 00:11:44,820 Image 2 143 00:12:00,330 --> 00:12:02,120 prediction 0 again 144 00:12:08,580 --> 00:12:15,030 the prediction is you mean this also is not a cut I cut into the cut into the solution. 145 00:12:15,030 --> 00:12:16,510 Okay let's see. 146 00:12:16,540 --> 00:12:18,030 There has to be something wrong somewhere. 147 00:12:21,540 --> 00:12:23,310 So let's try image 3. 148 00:12:23,310 --> 00:12:25,870 This is a major three major three say Cut. 149 00:12:25,890 --> 00:12:29,080 Let's try that already tried this. 150 00:12:29,220 --> 00:12:32,670 Let's try it 3 and see what we get. 151 00:12:32,910 --> 00:12:37,170 Image number three when I run over here 152 00:12:47,420 --> 00:12:50,810 this is a cut and the prediction is 1 1 means. 153 00:12:50,810 --> 00:12:51,570 True. 154 00:12:51,590 --> 00:12:53,910 So this was predicted correctly. 155 00:12:54,560 --> 00:12:59,330 Okay let's see image number for this car. 156 00:12:59,450 --> 00:13:00,220 Let's see. 157 00:13:00,290 --> 00:13:02,630 I'm gonna select image number four over here 158 00:13:15,280 --> 00:13:15,750 okay. 159 00:13:23,300 --> 00:13:24,860 Image number four. 160 00:13:24,910 --> 00:13:26,140 Prediction is zero. 161 00:13:26,920 --> 00:13:30,760 And as we know image number four is not a car. 162 00:13:30,920 --> 00:13:32,400 Let's to image number three again. 163 00:13:32,530 --> 00:13:37,420 I realize that the resort was out of the window as I was reading it to you. 164 00:13:37,480 --> 00:13:49,290 So I'll come back to major number three and show you what that prediction said. 165 00:13:49,960 --> 00:13:50,470 OK. 166 00:13:50,470 --> 00:13:56,890 It says 1 but have to reshape this. 167 00:13:56,890 --> 00:13:58,380 So we have it over here. 168 00:13:58,380 --> 00:13:59,660 Prediction is one meaning. 169 00:13:59,690 --> 00:14:00,430 It's true. 170 00:14:00,450 --> 00:14:02,100 It's a cut right 171 00:14:04,960 --> 00:14:05,930 and then. 172 00:14:06,370 --> 00:14:09,190 Let's see the next images image number five. 173 00:14:09,190 --> 00:14:11,910 Let's see what the car is going to be confused. 174 00:14:11,900 --> 00:14:12,580 That's it cut 175 00:14:16,100 --> 00:14:17,250 when I run 176 00:14:25,910 --> 00:14:26,350 okay. 177 00:14:26,360 --> 00:14:27,500 Prediction is zero. 178 00:14:27,500 --> 00:14:31,700 Meaning it's not a cut and this is correct. 179 00:14:31,700 --> 00:14:34,300 Let's see the last one you made. 180 00:14:34,340 --> 00:14:36,320 Number six is the bird. 181 00:14:36,380 --> 00:14:38,470 Let's see what a prediction is going to be. 182 00:14:38,480 --> 00:14:54,650 What does going to be confused for a cut. 183 00:14:55,590 --> 00:14:55,890 Okay. 184 00:14:55,920 --> 00:14:56,870 This is zero. 185 00:14:56,880 --> 00:14:58,180 This is not a cut. 186 00:14:58,330 --> 00:15:01,330 So it seems our new network. 187 00:15:01,540 --> 00:15:13,620 This work and well it it performs better find in non cut then um then find an actual cut mean meaning 188 00:15:13,650 --> 00:15:19,620 if you show a picture that is not a cut it's definitely it works better by telling you this is not a 189 00:15:19,620 --> 00:15:20,260 cut. 190 00:15:20,380 --> 00:15:27,090 It s performance is better when the image is not a cut but when you have an actual cut it's performance 191 00:15:27,360 --> 00:15:28,870 sometimes is food. 192 00:15:28,950 --> 00:15:32,290 Sometimes you would see a cut and it wouldn't predict that it's an actual cut. 193 00:15:32,360 --> 00:15:37,710 And like I said this is not the best way to implement image recognition. 194 00:15:37,770 --> 00:15:39,570 This just experimental. 195 00:15:39,710 --> 00:15:41,090 We would see a better fit. 196 00:15:41,160 --> 00:15:49,020 We would see um we would have two versions two new versions of the same implementation detecting cuts 197 00:15:49,170 --> 00:15:50,540 versus not cuts. 198 00:15:50,700 --> 00:15:53,120 We would go and use a deep new network to do it. 199 00:15:53,130 --> 00:15:56,160 And then after that we'll try convolution on your network. 200 00:15:56,160 --> 00:16:04,020 But I would advise that you try this out and then gets more cuts images and try it out and also play 201 00:16:04,020 --> 00:16:11,750 with the um the lending rate as well as the number of iterations to see if its performance can improve 202 00:16:12,270 --> 00:16:17,490 and you can throw some other images of known cuts and see if you can get it to confuse something that 203 00:16:17,490 --> 00:16:18,980 is in a cut for a cut. 204 00:16:18,990 --> 00:16:22,310 So far we've not seen it but when we've seen it it won't. 205 00:16:22,320 --> 00:16:23,580 We have seen so far. 206 00:16:23,610 --> 00:16:29,170 Sometimes you give a cut and it says it's not a cut but we've not seen it tell you. 207 00:16:29,190 --> 00:16:31,680 This is a cut when it's not a cut. 208 00:16:31,800 --> 00:16:38,100 What I mean is we've not seen the neural network confuse a known cut for a cut but we've seen it confuse 209 00:16:38,220 --> 00:16:39,940 a cut for a known cut. 210 00:16:39,940 --> 00:16:45,240 So yeah you should experiment with it more but yeah this or the rest for this lesson if you have any 211 00:16:45,240 --> 00:16:48,450 questions just send me a message if anything at all. 212 00:16:48,450 --> 00:16:54,150 If you have suggestions or you know you just want to run TiVo or something then or you just want to 213 00:16:54,150 --> 00:16:58,590 say something which regards the course you can message me and I'll see you later. 214 00:16:58,590 --> 00:16:59,190 Have a nice day.