Intro - TensorFlow Object Detection API Tutorial p.1

Key Takeaways

what is going on everybody and welcome to a tensor flow object detection API overview slash tutorials so in this video all we're going to be doing is just doing it just a quick setup of the tensor flow object detection API I'm going to try to I'm going to be doing this on Windows mostly since my main focus of using the central other detection API is in games which run on Windows and not a boon to and not very well as often on Mac so anyways I'm going to be doing on Windows I will point out a few slight differences not a boon too but it's basically the setup is the same so we're just going to set up get to the point where we can detect things in an image like this and then in the subsequent videos we're going to show how you might customize this to your own images and then to your own video and for my purposes the video will just will use you know streaming video from say Grand Theft Auto so anyway but you could change that to be your webcam or whatever and actually maybe first I'll just do it with the webcam anyway enough on that let's set this up so they actually do have installation instructions and you could totally just follow those there's like a few things that will hang you up though like for example like installing the potential flow GPU is not that simple it'd be great if it was that simple it's not so anyway so if you need instructions on installing tensorflow GPU I've got tutorials on that yeah so anyway one thing to note that's actually the reason why this is so useful is detecting things and images has all has been like something you could do for a very long time but detecting objects in images that you could stream those images like on a video so like to detect things at like 15 20 frames per second especially if you want decent accuracy is was a big challenge so and to write the code that's going to do this for you yourself which is take forever so this is a huge offering from tensorflow so thank you for that so anyways let's go ahead and grab this stuff so so I'm going to kind of assume that you do have potential already and so again if you don't I've got Stroh's on that also make sure you basically have all of this you can I mean you install it this way but I would just install these dependencies this way like pillow el XML Jupiter MATLAB if you're on Windows obviously without the pseudo if you do need admin just open up CMD as an atom so the only thing that might hang you up on your installation on a boom tube is if you're not on 1604 so they say 1604 and it's true but if you're on like 1404 for example like if you're using paper space the people learning in a box that's on a boom tube 1404 the only thing that will hang what there is basically you're put above compile compiler and I will cover what you'll need to do to fix that so anyways make sure you have all these including Jupiter pip install Jupiter and then once you have those I think I'll take it from here so so first thing we want to do is is get the models directory I'm going to just clone as a zip if you're on a boom tube just get clone github.com slash central as those models don't get super simple while we wait on that download I'm going to come over here it's just github.com slash google slash protobufs slash releases and these are all the releases for the protobuf compiler I'm going to because I'm on Windows grab this one so 4.32 save and if you need it on Mac for whatever reason you could download that here if you need it on a bin so you can download that here in the process will pretty much be the same at least on a on on a boon to you would just extract that zip folder just do sudo dot slash configure so to make or sudo make check and then sooner sudo make install on Windows if you consider you don't really have to do any of that to standalone so anyway I'm going to move this over here so let's go ahead and extract here the models and then while that's running whoops OH move it and see well that's running I'm also going to extract here Pro TOC which is just going to slap it into bin here now I'm going to leave it there just because for this tutorial i would suggest you probably take it and put it in like your c program files or something more useful than in downloads at least if you're like me I clear out downloads relatively frequently and so eventually I would just delete this by accident but I'm going to leave it here and then we just specify the full path so pro TOC is right and bin so yeah so moving over here once you have those two things we're going to come back over here and pop into object detection so the next steps are to setup that you know point that compiler basically so let me just come back over here installation so it wants you to do the protobuf compilation so from within that models directory I want you to run this now again if you're on a bin too and you run this and I figured the area's but if you get an error you probably outdated Pro TOC you can check it by doing just just in the command line your terminal pro talk - - version if it's like 2.4 or something that's the outdated version I believe the most recent is 3.4 yeah so if it's not 3.4 and you get in there go ahead and update it that's probably your problem so anyway so we need to do this step as well on Windows so I'm going to pop this over here inside models master I'm going to go and rename this is just a models that's what it is and inside just straight up models you want to run the following command so let's just go ahead and copy that come over here CMD right click now Pro talk isn't going to you're going to get this pro tox now to recognize blah blah so a lot of people ask me about like this is like something to me like I'm getting this unrecognized command well it is not in your path we could add it to our path if we wanted or I'm just gonna specify the full path to pro talk where did we go there we are I hope it's big enough for you guys to see I think it is anyway C colon slash users Saturday I think it goes users H and it was in downloads downloads slash bin slash proton I think that's right oh it was right there I'm blind anyway yeah so we run that no errors great we're good to go then if you're on a boot - or probably Mac as well you're going to want to export here to be honest I don't even know what that lines doing someone comment below and tell me what the hell that line does I'm going to wager that's not necessary on Windows and then finally you could test the installation but we're going to we're just going to we're going to run the via the Jupiter notebook so now we want to do it let's go ahead and change directory into object detection and now I'm going to do Jupiter notebook and that'll pop up the Jupiter notebook for us come down where's I paint there we go i pine notebook and this is just a very quick demo basically what it's going to do is it's going to download in fact let's just run all and I'll kind of explain what it's going to do so run all what's going to do is it downloads a pre-trained model for you and the pre-trained model is this one here which is coca which is stands for comp common objects in context Barbic alright I think it's 300 objects but I'm not positive let's just a bunch of common objects in context so anyway it's going to download that it's a pre trained model now later you could train it on your own you can basically train your own model if you wanted or you could also train like you can continue to train they're already pre trained models so depending on what your goals are one might be easier than the other especially if you're like trying to detect similar like things it's probably it's pretty much always better to go off a pre-arranged model so it's done so this just runs a couple quick examples that on loads of images you got a dog and a dog and then you've got that image that was on there on their homepage basically so running this whole ipython notebook takes a little bit of time because you're kind of setting everything up but the actual detection process like this here like this loop basically like once you go into the session image can be anything this is iterating through just images in the test paths which just downloaded two images for you but this here this process this loop is from what I've found to be somewhere between like maybe 15 and 20 ish frames a second so you can stream really well so anyway in the in the next tutorial I'm going to show you guys basically how we can take this you know super simple demo basically and begin to modify this loop here to do what we want on our own kind of images or video from either webcam or I'm going to be using just video from like Grand Theft Auto but you can use anything one it's already a pretty good object detector but if you want to detect custom objects you'd want to train to run so I'd also like to cover doing that as well so anyways that's all for now if you have questions comments concerns whatever feel free to leave them below otherwise I will see you in another video