Print Companion Age with airSlate SignNow

Print companion age

so today we'll look at how to run age and gender classification model which uses convolutional neural networks and is pre trained in the cafe library deep learning library so this is the page you know for that particular model I'll be looking all this in the you know in the description the discussion will usually contains a companion article for pretty much all my videos and there you should find all the resources here if you look at it which is going to download the model and because we are going to be running this on a Raspberry Pi you know we don't have a lot of resources but the Raspberry Pi can can run this model although it's a bit slow it should be able to Don this model perfectly let me copy this address okay now here I'm logged into my Raspberry Pi I'm just gonna make correctly for this I'm gonna shift to the act and I'm gonna download this module you can unzip the module and see what it contains let's take a look at this folder as you can see it's got a bunch of temporary files which I will remove now so that they cost less confusion I am also remove what I remove the example image so we have let me also remove this item we're now left with this cafe model for age detection and this is the pre trained model these are the weights this is the file that contains the description of the model that is the different layers and what they are and how they connect and the same for we have two more the model description and the model itself the pre trained model itself for 4h so we have two files each for age and two files for gender now you might have noticed that there's another file called mean dot binary proto this this warrants a little bit of explanation so see what happens is when a drip drip learning model is trained there's a lot of training data and then it's impossible for all the training data to be very consistent for example in lighting right and then you don't want the deep learning model to pick up features that you don't want to pick up because of differences in lightning for example right so method you called mean subtraction is used and then it's basically used to fight this this difference in lighting for example in this particular case so let's say this model strained using the image and data set right and then there's obviously going to be differences between pictures of two different men of different ages but you know you you want the system to have a kind of a mean or an average value for each of the channels for example RGB in our case right so this this particular mean or binary proto file contains those neutralizing factors for each of the pixels now please understand that as far as open CV is concerned although you are using a cafe model you don't need Cafe installed on the raspberry pie itself because open CV is enough it's able to load models from many different deep learning libraries cafe being one of them and cafe is not going to be there along with open CV opens open CV is good enough to read the model having said that we need the values from this minted money photo file and if we are to read this file from any predefined model pre-trained model we will need Cafe installed and tossed me installing cafe on a system like Ubuntu it's not a joke and Andy it's even more difficult on on the Mac so what I've done is that I have installed cafe on Ubuntu VM I am switching I am switching to that VM and I have a file here I have a folder here which which has the mean dog binary proto and I have a simple Python script which I will show you now that's it it imports cafe and then it reads the mean ordinary proto file it parses it and it prints the three different values for r g and b so let me run this now and we have these three values okay and these are the three values we need from this right we have this and we are good enough we'll have this and we'll come back to this later on now back in a raspberry pie let's get ready to take care of this and write a simple script let me remove this is a file and rename this folder to something more manageable okay and I'm gonna open up a file and we'll start coding away now we need to import the Python PI camera module because we gonna read from the Raspberry Pi camera and we need the imager tools time and of course OpenCV which is CV 2 if you want help installing open CV on the Raspberry Pi I'll link to a video in the video description which you can refer to now let's go ahead and initialize the camera here we opening up the camera setting the resolution the frame rate and you know we setting a prog-rock capture to a particular array size now remember we got these mean values here we gonna store them so that we can we can we can use them as we start inferring from the model we'll sleep just a little bit to allow the camera to warm up then as the first step let's try and capture images and see whether we are able to display them right we will do that before we can we will go to any other you know area right here will be a main execution point entry point call capture loop start capturing frames from the camera and we grab the image from the frame dot Raa and this is where we use the hawk cascade prom will face recognizer okay to recognize faces and tell us where in the image the faces then we convert the image to grayscale and we asked the hawk cascade face recognizer to detect multiple faces and we pass the grayscale image into it into this function and faces returned will basically be an array of faces that the HAR cascade found alright so I think it's clear till now now we loop faces where we will be able to get the X Y width and height of the detected face itself and we'll draw a rectangle around the faces found there doing this with the opencv rectangle function so we are passing in the image the original image not the grayscale and we are you know we have the X Y and width in height where we want to draw the rectangle we show the image we wait for a key press for one millisecond otherwise we continue we clear the frame we clear the stream we check if we press the Q key and if it was indeed pressed then we break so let's go back to the sloop again we go we we set up a camera for continuous capture we get the numpy array in an image that is the the array which contains the image and then we set up a AHA cascade classifier for a full-frontal page full frontal face I'm sorry right here and this is the location of the Hart cascade that OpenCV instance we convert the image to grayscale and we pass it on to the Hart cascade and we pass it pass it in right here and then heart the heart cascade function returns multiple faces it can return multiple faces form we loop through those faces get the X Y width and height locations and simply draw a rectangle around the faces we show the image with the rectangle in a window and then we get if any key if we see if any key is pressed and if it is we will check whether if the key is q and if it is the Q is placed on the window we break which essentially means we get out of the main fot loop which will end this function and end in the script as well okay now let's run this and see what happens we are done with this Google to virtual machine so I'm going to shut it down and I'm gonna log in to my respective Pi in another terminal window so that we can easily run the program okay we are in let's run this and see I'm getting into a virtual environment which contains OpenCV if you don't know how to set this up please refer to my installing open CV on Raspberry Pi video which I'll link in the video description I forgot to pass the X & Y options which let me remotely open the window I'm gonna do that now you should see my face with the rectangle around it let's check that there you go a la I'll press the Q key now to get out okay so we have our basic script running we have what we have let's go back to the script and what we have done here is that we set up a loop and that captures frames one by one we have the HAR cascade we're drawing a rectangle around it on any face detected face off faces and the script seems to work now what we need to do is that the idea is to around each rectangle we also gonna print the gender and the age of the face that is that is detected all right now the main thing here to understand is that when we pass an image of a face to the the cafe model the age and gender detector model we cannot pass the whole image you know as the camera captures it because the model wants a face closely cropped and facing fully frontal okay so what we are doing here is that using the hawk cascade we detect the full frontal face we crop it and we send the cropped image to you know the model to to predict the age and the gender let's start adding support required to get our age and gender model upper turning matting to lists this is one list which has the range of ages that the model predicts and of course two different genders I guess it's not gender-neutral yet but this model has at least made and female right now let's load the models mean now right before this loop the camera loop starts I want to load these models in Wilma screwing up all my pastes but I'm gonna fix that later there you go so now here we have our age and gentlemen gender models being loaded we are using the Reid net from cafe function which the DN and the deep neural network model the D electron module of open CV has right and as you can see the proto door txt which basically contains how the model is tagged and how the model is designed and the actual the the weights of the time model itself they are both lis loaded so these two files for each model are required carrying on now let me split this into another function so we have a function now which will return the loaded models which we will use let's say and pass it on to our capture loop function and we'll pass this to these two models to the capture loop function and bring these in as well into this function okay now when the script runs the capture loop function will have access to both these models now let's go ahead and see how to infer the age using this model from our names from the raspberrypi camera now what I'm going to do is I'm going to create an an image of just the face you know this the beauty about the Python OpenCV library is that we can actually just give two different slices of the image the region of interest we need and then that which we get from the rectangle which you know which is available from the HAR cascade we take that and we create the face image so whenever I whenever you see that there's a rectangle drawn you know around the face that recognize that rectangle or a square we make it available as a separate image and we pass that to the model rather than passing the whole image now this is the blob from image function which is there in the deep deep neural network model of OpenCV this is where we pass the the mean values which we have here on top right which we have here on top and this will essentially pre-process the image so that it can be passed on to the model for prediction let's start predicting the gender now let's carefully go through these three lines that predict the gender so we have a gender net which is the genders pre-trained model which got passed to our capture under school loop function this function we are currently editing we say set input and we plot past the blob object which essentially is the image corrected with the mean values right and then here's where the magic happens we say gender dotnet forward function which essentially predicts and then returns the predictions so this is a typical multi-class classification problem and what we have here is that a set of predictions are returned and here we have one two three four five six seven and eight different classes possible right and we will get returned an array which has basically integers that tell us what is the possibility of you know each particular class eight different classes so the highest for example let's say 15 to 20 turns out to be 0.7 it means that all the others are much lesser than 0.7 it means that we will have to assume that the model thinks the particular input the person in the input is between 15 and 20 years of age right so this is exactly what we are doing here so gender list is the is the list we just examined with the different classes so what we are doing here is that we are taking gender predictions and then we are saying Arg max so we are saying look at the items in this array and return the position of the item which has the maximum value right so it selects the right item in the label right so let's see some examples of that it's very easy to see when it's actually printed out right so if you print this out right now I'll go ahead and copy five of these lines because this is what happens to age as well all right so we have the code to predict H now let's form a string so that we can display it near the face of the person on the image now this will be our string we have gender and age for example it will say male or female , the age range right and then we have the we use the CB put text function which essentially displays text open Seavey's way of displaying text we are missing the font here which I will now copy let me put this outside the for loop all right this will be a font okay we save this now we could try running this okay we are the wrong project I guess a gender model deploy age is the problem here son able to load this okay you specify the name is models here correct that and go back clear the screen okay there you have it it's pretty confused about my age also things anyway let me reduce the font size a bit going back on scale you wanna reduce that a bit and done it again okay there you have it make with this now I wanted to focus on these two outputs this is the prediction of male and female so it thinks this is the possibility of the person being male and the possibility of the person being female and this is the different possibilities of the age ranges right so because we're using the function here which says arc max it's going to take whatever is the maximum and we are using that index to select one of these items from the list right so this is how you know you can use cafe models in open CV it's it's very very easy to understand and it's pretty powerful because although you cannot train very successfully on the raspberry pi you can use very powerful deep learning toolkits and create models pre train train models and take those pre train models load them on the open CV and run them successfully right and it doesn't stop there there are other deep learning models as well as an exercise you can you can go and find out what other deep learning toolkits are supported by open CV right that's one thing you can do and of course it's a little bit slow because you know it's running on the CPU number one and number two single core core performance CPU core performance on the Raspberry Pi it's nothing to write home about right but nevertheless it's a very interesting idea and hopefully you will get a lot more excited and check out all the other preteen models available there are lots of pre train models available and you can do many many interesting things with open CV and pre train models from other deep learning frameworks I hope you enjoyed this and see you until next time bye bye