Google Teachable Machine Learning Build No Code AI solution

welcome to own little coder in this video we are going to see how to create a machine learning model that uses computer vision to detect whether you are touching your face or not so in this time of coronavirus reading so this is a project that was initially initiated by the Kaos founder on Twitter and that tweet has given inspiration to a lot of projects one of the projects that I recently came across is bike race Kumar and his project is based on PI touch and Yolo v3 so in this project we are not going to use any programming but we are going to use the training images that he has collected so thanks krazee Kumar for collecting the training images for us so in this video what we are going to do is we are going to build the machine learning model without even touching a single line of code and that is we are going to do using teachable Machine with Google so Google has some time back created this project that will help you teach a machine and it also gives you the output model in a variety of ways so if you want it to be a tensorflow light for your edge devices you can download it if you want it to be at inflow tensorflow dot JS for a website so you can download it or if you want the normal tensorflow file you can still download it so what we are going to do in this video is we are going to use the teachable machine web interface and we are going to build a machine learning model that will ultimately help us detect if someone is touching their face or not with the picture so this is just going to use a picture but you can do the same thing with the videos just you have to have the respective training dataset so let us get started first and then see what are all Google is asking us so so it's an image project and you can see that you need to have class one images class two images you can use your webcam which is not I'm going to do but in this case we are going to use static images that we have downloaded from internet so once we have two class of images then we can train our model and then we can export our model so let us go ahead and see our images so some of the images that I'm using is downloaded by Chris comer and some is downloaded by me and we just used Google Images for it so first we are going to have a folder of images where it is all about people touching their faces so if you want I can just increase the size yeah so it's all people touching their faces different images and then we have got the other set of images where people are not touching their faces this is quite less but it's okay so this is just for demo and we are going ahead with that but if you want to put the model in production please make sure that you have got enough training dataset and we are going to test it with two images one is where know the person is not touching the face the other one is where the person is touching the face so given that this is our own training in one class and training second class and then their test we are going to get started with our model building activity so let's call the class one face touch let's call the second one node touch and then let us upload the images from you have the ability to upload it from Drive but here we are uploading it from local images so we are going to go to the place where we have got our images the more images face touch select all open then the neck and second classes no touch upload from images go back no touch select everything open so now that we have got down both the classes one is the face touch on a snow touch let's train the model training the model shouldn't take a lot of time in this case I think when I tried it to come few minutes let's see how long it takes in case if you are thinking that if this is just a by classification option you can have multiple classification multiple class for your target variable you just have to add a class so it's it's just asking me if I want to although my webcam something don't do it so I'm just going to use a file and in this case or modular successfully trained we have options to export our model but before they let us see how the model is working so click the image nah we don't want to touch anything from our training so just to let you know that none of the testing images that we are using is from not our training ok so the testing image is completely to random image so let's let's pick this guy without touching the face so it's saying 81% not touching the face so that is good so our machine learning model is working fine so let us try the other image which is the person touching the face and it's saying 100% the images are touching the face so if you want we can go to Google and then try one more image so let's say person not person face and let's let's download this image sorry um this image looks nice so let's let's download this image we have a lot of images image - instead they say demo images just let's pick one person touching face oh which one is better so in this case let's let's - let's take this one save as frustrated man today I am I just just okay okay I didn't know where did I save it so I'm just going to save it again yeah it's there it's there so let's go back to our testing and go to the images that we wanted to select so let's pick this man who doesn't oh it's an invalid image type okay so let's let's just select this man and see what happens man touching face it's it's it 100 was introducing at the - touching the face [Music] man version phase let's let's try to the girl just kiss this one and let's go to a model in the image okay um this this shows how the model is not doing good this game shows why it is very important for you to train your model very well and again you have a lot of hyper parameter tuning available for you say you can increase their learning rate or to speed up the process you can increase their poke you can increase the batch size you can play around with these data in order to it and play around with these parameters in order to change the model building process but I think you get the gist of it the objective of this video is to introduce you teachable machine and then how can you train a machine learning model for computer vision problem especially computer vision problem without touching a single line of code um so once you are done with all of these things um so let's let's let's pick one final image and then see how do you majesty sorry I'm with the same image again oh it's 52% so model is doing decent it's not worse but it is doing decent but yeah it requires a lot of fine-tuning more training data before you and we move to production so this is only for educational purpose so how do we export the model click export the model so you get a bunch of options so there are two ways one you can download your model locally second you can let teachable machine on website to host your model on cloud or in this case you can just refer it online so you have three ways to download your model one assort ends upload or jas second one is tensorflow where you can use it with the typical Python script if you want to do the automation or Raspberry Pi or whatever it is so tensorflow not J's if you are going to host it on a website attention flow light if you want to use it on Android device or any kind of edge device so it gives you a T of light so yeah that's that's the main thing of this video where we picked up some more training images and then we built a model that detects whether someone is touching their face or not so in this time of crisis don't touch your face and maintain social distancing and I hope you stay healthy and also have some fun with machine learning thank you for watching this video see you in the next video