Cloud OnAir: AutoML vision: Making custom image analysis possible for every business

Cloud OnAir: AutoML vision: Making custom image analysis possible for every business

Show Video

Hello. And welcome, to Google cloud on-air these, are the live webinars, from Google cloud today. We are hosting webinars webinars. Every Tuesday my, name is Elias Pinto and I and with me I have I'm, Sarah Robinson we're. Based out of New York and today. We will be talking about Auto ml vision. To. Let you know week you can ask questions anytime, on the platform, and we have Googlers, on standby to answer them so, let's get started so here is Sara thanks. Louis so. Today we're gonna talk about Auto ml vision making, custom image analysis, possible, for every business. Before. We get into that let's talk about what machine learning is so. At a high level machine, learning, involves teaching computers, to recognize patterns in the same way that our brains do, over. Time as machine learning models are given more examples, and experience, they're, able to improve and generalize, on an input that they haven't seen before. So. Does anybody remember how they learn their first language, her. Parents probably didn't give you a dictionary and a bunch of grammar books to memorize that'd be kind of weird instead. You learned over time by being exposed to many different examples so, let's say you were having pasta for dinner you, saw it on your plate you heard your parents identify it maybe, you identified, it incorrectly, a few times and were corrected, over. Time this repetition strengthens, certain, pathways in your brain and this, is roughly how machine learning works - so. At a high level machine. Learning is loosely, based on how the human brain learns, instead. Of biological. Neurons we have mathematical neurons. That mimic these neurons in our brain with. Machine learning we. Can solve problems without, exactly knowing what, the solution might be and. Finally machine learning enables systems that, improve over time as they're given more and more data. So. Before I start solving. A machine learning problem and picking the tool I'm going to use I like to think about the type of ML problem, that I'm solving and, on. Google cloud platform we, have a spectrum, of machine. Learning offerings, so on the left hand side we have products that are targeted more towards application. Developers if, you're new to machine learning and don't have any machine learning experience, we, offer five pre-trained, API, these give you access to pre train machine learning models all you need to do to access them is make, one REST API request, on. The right side of the spectrum we. Have products that are targeted more towards data scientists. And those with a little bit more machine learning experience. And for this we have tensorflow for building your own machine learning models an ml, engine adds our managed platform, for tensorflow. So. Going a little bit deeper into this if you've got a common, classification, task that someone else has solved before you don't need to start from scratch you don't need to build a model from scratch trained, only on your own data you, can utilize existing, pre, train models that are out there but. If you've got a problem that's very specific to the type of data that you're dealing with at your company it's.

Probably Going to be a more custom task that you'll need a more customized, solution for. So. Let's take image classification as, an example let's. Say you want to identify that, there's a cat in this image you've. Probably heard this example before this, has been done many times before there's lots of models that exist out there that have been trained on hundreds, of thousands of images that, can tell you if there's a cat in an image or not so, there's no need to start from scratch although you can if you'd like but. Fastest, to utilize a pre trained model that's already out there to help you with this but. Let's say for example that this is your cat its, name is Chloe and you want to identify Chloe. Apart, from other cats in your image in your image data set so. This is a more custom task you're gonna need to, train, a model on your own data, using. Label data so that it can identify Chloe. Apart, from other cats, and animals in your image, library. So. Let's start with the machine learning api's, we offer five, different api's, that. Give you access to a pre trained model to accomplish, common, machine learning tasks, so, you don't really need to know anything about how, these models work under the hood to get started, you just pass it your image data for the vision API for example you get back a classification, the, video API does what the vision API does for images but for videos lets you analyze videos, the. Speech API lets you perform, speech. To text transcription, with. The natural language API we can further analyze that, text and with, the translation, API we, can translate text, into over a hundred, different languages, so these are our pre trained api's targeted. Towards application, developers, who want to integrate machine, learning in their applications. On. The, other side of the spectrum we, have tools to help you build and train your models from scratch so. From the beginning the, Google brain team wanted everyone in the industry to be able to benefit from all of the machine learning product, projects, that they were working on so, they made tensorflow an open-source, project on github and the, uptake has been phenomenal tensorflow. Is the most popular, machine learning project on github last. Time I checked I think it had over 90,000. Github, stars it. Also just crossed over a million downloads it's, being used all over the world and because, tensorflow is open source you can train and serve your tensorflow models, anywhere. Because. We're focused on Google cloud platform today, I'm going to talk about our managed platform for tensorflow this is called cloud machine learning engine and it. Lets you take advantage of distributed, training you can distribute, your training. Job across machines using, GPUs. Or TP use to accelerate, your machine learning workloads once. Your model is trained you can then choose to deploy it to machine learning engine, for serving so. Let's say your ml model becomes a huge hit you're getting thousands, and thousands of prediction requests every second, you're, gonna need a scalable, way to handle all of those requests, when. You can use machine learning engine, to serve your model a great thing about ml engine is there's no lock-in, you can choose to train your model on ml engine and then download it export. It and serve it somewhere else optionally. You can train the, model wherever you'd like and decide you want to serve it on machine learning engine. Another. Thing I like to think about when I'm diving into a machine learning problem is the resources, that I'm gonna need to solve that problem I've. Listed just four here there's probably others that I haven't thought of but. Let's go over what these are so first is training data how much training data are you going to need to. Get an accurate model. That's going to get an accurate prediction on your data how. Much model code are you're going to write. How. Much training and serving infrastructure, are you gonna have to provision, and then finally overall how much time is this whole process going to take you so. Let's look at how this applies to the products that I've talked about so, far. So. If we look at machine learning as an API the. Great thing about these API is is that you don't need any training data to use them the model has already been trained they're pre trained models so, if I wanted to I could just generate a prediction on a single image I passed that image to the API I get, a prediction back but, it didn't require any of my own training, data I also.

Don't Need to write any of the model code again that's all handled for you by the API, and. API is available on Google cloud platform so, I don't need to worry about training or serving infrastructure, all, it's gonna take is a little bit of time to, write my request to the API and figure out how I want to parse the response that I get back you. Can get up and running with these api's in probably, less than a day. Let. Me talk about two approaches, to building a custom model the first is transfer, learning which, lets you take advantage of a model that's already been trained to do a similar task to what you'd like to do so, let's say you want to identify where certain objects are in an image there's. Lots of models that have been trained to do this so you can utilize the weights of these trained models and then, just modify, the last couple layers based, on your own training data so. With this you will need some, of your own training data because the output is going to be the, predictions are going to be specific to your data set you. Will need to write some of the model code and you, may need to think about provisioning. Training or serving infrastructure, or you may choose to use a managed, service for this and. This will obviously take a little bit more time than, the pre trained api's and. Then. Finally, if you've got a model built entirely, from scratch trained, only on your data you're, gonna need a lot of training data for this to work successfully. You're. Also going to need to write a lot of the model code yourself, think. About where, you're gonna run your training and how you're going to serve your model in production, and obviously. This one will take even, more time than with transfer, learning. So. Focusing specifically. On image analysis, let's take a look at the existing tools for image analysis, on Google cloud platform. For. That we have the cloud vision API which, lets you perform image, analysis, with, a single, REST API request. These. Are all the features that the vision API version, API provides so, at its core the vision API offers, label, detection which will tell you what, is this a picture of so. For this image of my return elephant, animal. Etc. Then. We have web detection, which, goes a step further and, it will look across the internet for similar images and then, based on the content, those pages it will provide additional details, on what's your in your image, then. We have OCR, or optical character. Recognition, OCR. Is, able to extract text from images so if any of you have ever used the Google Translate, app before and taking. A picture of a sign and then seeing the translation, in real time you, can use the vision API OCR. Method to implement similar, functionality, into your own applications. Logo. Detection, will identify common logos in an image, landmark. Detection will find landmarks in an image crop. Hints can help you crop your photo to. Focus on specific subjects in the image and then, finally we have explicit, content detection, this, is pretty self-explanatory, it, will tell you it's, your image appropriate, or not this. One is pretty useful across the board for any any site or app that has user-generated content, instead. Of having someone manually, review all those images that come in you. Can send it to the vision API and, then maybe you only need to review a subset, of those images. As. I mentioned the vision API is a REST API so, you can call it in any language you'd like the. Example code I have here is in node J s and on. Google cloud platform we, have a number of client libraries in. Your favorite language that. Make it easy to call all. These api's, so i'm using the google cloud node module for, the vision API, here. I tell it the types of detection I want to run I call. Detect on that vision object, passing, it I can either pass at a local image URL, or, I can, pass it the URL of a file stored in Google Cloud storage I can. Also just pass it the raw image content basics before encoded, so I pass it that image the. Types of detection I want to run and then I get a response back with the detection x' and i can analyze that response however I'd like to use, that data in my application. So. I don't like to get too far into a presentation without. Showing a live demo so I'm gonna bring Elias on here to show us a, demo. Of, vision. API Thank. You Sara so, here. We have a chance to actually see how powerful the, the cloud vision API really, is and and, to build on what Sara just said it's really gives. The ability to understand, what's in an image it's and Google is very uniquely. Positioned. To be. To, provide you with powerful machine learning models and let.

Me Show you how easy it is this. Is a publicly available, demo. That you can do and we're. Gonna go and actually pull up some. Images here, so. Sarah was kind enough to share some of her recent. Holiday. Pictures and we, pulled up a picture she took in Poland. And right. Away what, you see is that the image, the. The machine learning model immediately. Recognized, that this was taken, in Krakow, or a few polish as I think you pronounce a crack off and and. It, not only recognizes. That landmark, but it's able to find. Out that it's also the main market. Square and because. Of Google Maps were able to pull exactly, the lat/long, of, where. That picture was taken, let's look a little bit more into the other labels. That we have the other entities that we recognize so, we've recognized, the face see, here we see the, human faces is picked out but also please note the, exact, location, of things like eyes nose. And mouth. Is correctly. Identified we're. Able to also determine. A, sentiment. Analysis, so, Sarah. Here looks very joyful and even surprised at that picture and. As, you can see they're very likely those. Are the emotions she is displaying. On that picture. Must have been a fun holiday, so. We can quickly see there is a very long list of of. Labels, that we automatically. Assign. To that, picture and, we're. Able to see that it under. The the. Web, entities. We're. Able to use the knowledge of the, web about that picture and actually pull it and does, an image search and compares. That with the information, available on the web, I'll. Briefly mention here, that we, also have safe search we, can very quickly identify. That yes this is just the plain holiday, picture. And it's, safe to show at any age last, thing I want to show is the JSON output it's. A very long output out but I'll just highlight some, of the types. And positions, of the different labels so you see the left eye the right eye and. So on and so forth for each one of the labels that it identified. It pulls, it, gives you exactly, where it is located on the image, let's. Pull up another, image and this. One is a little busier. Sara. Again was very lucky to have gone to Hamilton, the show very recently so she shared a picture as you, can see this is a much busier picture there's a lot more going on a lot more different labels and, and and and and and other. Other. Elements, in this picture under. Faces right away we picked her face here again with a sentiment, analysis, we, want to show now, here's the interesting bit, we. Can show things. Like on. The web it recognized, correctly this was the Richard Richard, Rogers theatre and under. Texts I don't, know if you notice but the word playbill is actually in the bottom there this highlights, the intelligence. Around OCR. That we have even if there's multiple words in the picture we're able to extract that to you if it's a piece, of text, under, the document, we. Say, it's a menu or, a, form, or a. Resume. We're, able to pick different blocks. Of, text. And display them on their pages and blocks so. This. These are some of the things that we wanted to highlight the good thing is that this is publicly. Available under cloud Google, com slash, vision, and you may test this on your, own today. A quick, reminder if I may if you want to ask questions we ask, you that you put. Those questions throughout our presentation so. We're able to bubble. Them up towards the end of our session back. To Sarah Thank You great. Demo. So. Back to the slides I want. To revisit the ml spectrum, that I covered in the beginning so. Remember that on the left hand side we have products, targeted towards app developers, machine, learning api's these, are pre-trained, api's you, can't customize them at all but they're very easy to get up and running with on. The other side of the spectrum we have products targeted towards data, scientists. And machine, learning practitioners. And these allow you a lot more flexibility, in building a custom model train, on your own data set so. This is what the spectrum looked like until, just. A couple of months ago I'm, super excited about this, new product. That we announced, in January, called Auto ml the. First product under auto ml is Auto ml vision it's. Currently in alpha so you need to be whitelisted to use it but what, auto ml does is you can see it clearly fills a gap in this spectrum what, it does is Auto ml vision lets you customize the, vision API to.

Your Own data set so let's say the vision API works, pretty well for you. But you've got some very specific images, that you wouldn't expect it to recognize what they are but. You want to generate labels, based on your company's very specific, data set now. You can do that with auto ml. So. This is what auto ml vision provides it's essentially, a UI that helps you with every, step of building, your model from importing, the data, labeling. It training, a model, evaluating. How well it performed, and, then finally, making predictions, on that model using. Data that it hasn't seen before so. The best way to show you auto ml vision is to. Jump, right into a demo, so. For this demo let's say that I am a meteorologist. Maybe I work at a weather company and. I. Want, to predict weather trends, and flight plans from images, of clouds so. This begs the question can, we use the cloud to analyze, clouds the. Answer as you probably guess is yes. So. As I was preparing this demo I didn't, know much about actual, clouds before so I learned a lot and I learned that there are many different types of clouds as you can see here and. They. All indicate different, kinds, of weather patterns. So. My first thought was to try this out with the vision API I took a bunch of these different cloud images and send. Them to the vision API to see what I got back now. For us it's easy to see as humans, that these are completely different types of clouds but. The vision API was trained on a huge variety of labels to recognize. High-level things so it knows that there are clouds in these images but. Even though these images are very different types of clouds we, get similar labels back from the vision API, so. This is where Auto ml vision comes. To the rescue so let's see a demo of auto ml vision go, back to the demo and what. We have here. Is the, UI for auto, ml vision so as you can see it, takes us through each step of building, our model from importing data label. Train evaluate. And predict, so. The first step here is to, import our data and there's. A couple ways to do this we can put all of our images in Google Cloud storage and then, we need to create a CSV where the first column will be the URL of that image the, second column will be the label associated, with that image now. We can also build models that, have multiple labels for images for, one image then you just add extra columns to the CSV for that particular, image, so in this case I've already uploaded, all of my data and train the model so, let's see what happens here in the labeling, step, so. Here is where I can review, my image labels, so. I can see what I've assigned each image this, data set I've already labeled so this for example is a cumulus, cloud we've. Got a cirrus cloud here, let's, say I label this one incorrectly, I can click on it and switch, out the label here. But. Let's say for example. That I do not have time to label, my dataset or I had a giant data set and there was no way I'd be able to label, at all Auto, ml, vision provides, an in-house, human, labeling service so. The way this works is you provide a set of instructions, and some exemple, are images for each label -, these human labelers and in, just a couple of days, you'll get back a labeled, image data set. For. This case I already had a label data set and just to, look at a breakdown of how many images. Of each type of cloud I have here I'm looking at five different types of clouds and we. Can see how many how, many images I have for each label you. Only need ten images per label to start training a model but, Auto ml recommends, at least a hundred for, high quality predictions. And you'll obviously need more depending. On what you're trying to predict. So. The next step is training, your model and the great thing about Auto ml is that you don't need to write any of the model code that's handled by Auto ml for, you automatically. All. We need to do is literally, just press this train button and we, can choose whether we want to train a base model or a more. Complex, advanced, model I'll get into that a bit later on so once, I've trained my model I'll get an email when it completes, and my.

Next Step is to go in here and see how, my model performed, there's. A bunch of different machine learning metrics, here, what, I want to focus on is what's called a confusion matrix if. It looks confusing it's called a confusion matrix but, it's, actually not that confusing let's take a look so in ideal confusion, matrix we want to see a strong diagonal, down the top left which is what we get here what. This is telling us is when. We uploaded, our images to Otto amel it, took a portion of them to train the model but then it set aside a, small, percentage, of those images to see how the model did on data, that it didn't see during training so, this is how the model performed, on that smaller test set of images so what this is telling us is that for all of my images that, were actually, altocumulus clouds. My. Model, predicted. 76%. Of those correctly, and we, can see the percentages, we get all down, here now. For this model I actually trained, both the base and the advanced version and auto. Ml has a great way to compare, different versions of your model so, let's say even if you just train a bunch of versions, of the base model let's say you train it once with 500 images then. You add 500, more images and retrain, your model and you want to see how it, performs so I can compare them here. And now I, can see how, it did on the base compared, to the advanced model and this. Looks pretty good it looks like this. All of these went up significantly, when, I went, from the base to, the advanced model but. Hey wait this one actually went down 12%. And wouldn't, you expect the advanced model to perform better, across the board well. What this actually points out is that, there may be some problems with my training data so. If we look at where. It got confused, it looks like it was labeling, a lot of my altostratus images as cumulus, clouds and I, can actually click on this and I. Can see the images that my model was confused, on so it turns out that these images actually are a little confusing and they may have been labeled incorrectly, and. The advanced model was, able to identify. These shortcomings, in my training data for me so. Now I can go in next time I update. My model I can make sure to add better. Images, of altocumulus clouds. So. The next step and the most fun in my opinion is generating. Predictions, on our train model and there's a couple of different ways to do this, one. Is I can use the web UI so that's what I'm going to do right now I can. Upload an image here and I can see what. Auto ml vision is going to predict. So. This was indeed a picture of a cirrus cloud and this is pretty awesome Auto ml is our model is 99% confident, that, this is a cirrus cloud so. We can use the UI to generate, predictions this, is a good just quick and easy way to check your model see how it's performing right, after it's been trained but. Chances are you actually want to build an application that's. Going to query your train model and. There's a couple different ways to do this I want. To highlight one, in particular, which is the vision API. So. You saw briefly in the past slides what. A request to the vision API looked like and what, I want you to focus on here is that the, request really, doesn't look that much different with. Auto ml so the only things we need to add to our request is this custom, label detection parameter and then. Once we train our model we, get access to this ID for, our train model and so we need to add that to our request only. I have access to this model version or whoever else I share this project with, so. Let's say that I'm using the vision API let's say I had some weather app and first. I wanted my app only, to detect, is there a cloud in this image I was using the vision API but. Then I decided you know what that wasn't actually good enough I need to detect the types of clouds that I'm seeing in these images I don't, really need to change much about how my app is architected, in order for that to work I just need to switch out or.

Add This custom label detection parameter, along, with the model that I've just trained so. That's how we generate predictions on our model and. To show you how easy it is to build an app that uses our train model I've, built a super simple web app here where, I can upload a photo of a cloud and we'll. Use this one and it's gonna call my model I'm. Just gonna tell me a little bit more about that cloud so. This is a cumulonimbus, cloud, and if. You see this type of cloud while you're on a plane probably, not the best sign might be some turbulence so. This this is just a simple web app that queries, that model using the vision aid which I just showed you and it, gives us a little bit more data on what, these types of clouds actually. Mean so. That is Auto. Ml vision gonna go back to the slides. And. I wanted to highlight some companies, that are using auto ml vision in production, the. First is Disney Disney. Built, a model train to identify specific Disney, characters, products. Categories, and dominant, colors that were specific, to the images that they have in their data set and they've. Integrated this model into, Disney search engine and, it's, helping their users get more relevant results and find. Their products faster. Urban. Outfitters is a clothing, company and, they, have a similar use case to Disney they train a model to recognize specific, types. Of clothing patterns and, neckline, styles and they're. Using that to build a comprehensive, set of product, attributes, so, through. This they're able to improve product, recommendations, search. Results, and product filters, the. Last example, is the. Zoological, Society of, London, Zoological. Society, of London has cameras, deployed all over in. The wild to. Identify, different types of wildlife. In their images and. Rather than having someone manually, review all the footage that they're seeing they, built an auto ml model to identify all, the different types of wildlife, that they're seeing in those images. So. I definitely encourage you to go learn more about Auto ml vision you, can go, to, slash. Auto ml and there, you can fill out a form to sign up for the alpha we're, really interested, to hear what types, of use cases you have for auto ml there's. More details in the blog post where we announce Auto ml there's. Also an intro video a podcast. That goes into a little bit more detail about how this works under the hood and. Then if you want to try out the vision API if, you want to use the demo that elias showed earlier you. Can head over to, Slash. Vision to try that out with, your own images I definitely recommend doing that let's, say you want to see if the vision API is right, for your app but. You don't want to write any code yet so you can just upload your images there see, the JSON response you'll get back and. See if the vision API is, going to be a, fit for your, application. So. Thank you and. I'm. Gonna bring a line us back up here so we're. Gonna be going into Q&A, here in just a moment so we were back in a minute, thank. You. You. Welcome. Back everybody hopefully you've been busy, putting some questions there and we've, bubbled up a couple of the questions here to get us started so, do. You want to take the first one here I'll read it first so how, do I sign, up for Ottawa now that is a great question to sign up for auto ml go to cloud Slash Auto ml and there, you'll see a link to fill out a form and this. Is where we're collecting details about the, use case that you want to use Auto ml for so we'll. Review that and. Then you should hear back within, a couple of weeks or so. Let's. See the next question so, you want to take this one sure what. Industries, do you see using Auto ml the. Cool thing is that we are democratizing. The technology, so really any business. That has that. Deals with. Image. Data and which, arguably, is the vast, majority whether, it is actual. Image. Data that has and specifically. I'm talking about Auto ml vision here if you have to as part of your business identify. Things if, you want to and. We talked about Disney, about, identifying, their their characters, in that intellectual, property across. A very large data. Set we're able to. Really. Apply the. Mature. Machine. Learning models here to a lot of verticals. Like manufacturing. Retail. Any. Sort of, customer-facing. Applications. They're. They're way too many industries. Here to to I don't want to limit ourselves because, even.

If You're just processing. Resumes. Or or, credit card I'm not credit cards or business cards we can do a lot to. Expedite. What your business does. Lower. The costs of entry because again it's a it's an API you, don't have to build another, Google, Data Center we've done it and so you're able to just leverage that and and. Just pay for what you use. Okay. I'll. Take the out let me read the next one so how how. Do I know whether I should use auto ml or build. An image classification map, model, from scratch what. Is that one yeah that's a good question so. What Otto amel excels at is image classification so, you. Give it a lot of image labelled images it's, able to output a label, or tag for. That image but. Let's say you want to do something a little bit more complex, like you want to identify. Bounding. Boxes in your image so you want to say this. Is a specific, product. That. Our company builds and we won't identify where we where we're seeing it an image with. The bounding box you, would need to build a custom image classification model, with, tensorflow or, cloud machine learning engine to, do that or let's say you wanted to identify. Regions. In an image that's called masking, so identify, labeling. Specific. Regions, and whatever, types, of images you're labeling, that's. Another use case for a, custom. Image classification, model, but. If you if you would like to do image tagging, so output some labels based on an image auto. Ml is great across the board and, the, really cool thing is it doesn't require you to have, to write any of that model, code yourself. Awesome. Let. Me read you the last one the next one here, how long does, trainee a model. Really take and are, there any best practices. To consider, in order to speed up this. Process or, improve, accuracy. Sounds. Like somebody's really really. Wants to train their model yeah, that's. A good question I can take that one not so. How long does training model take for. The base model. I found, it. Takes about an, hour maybe a little bit less, the, advanced model for, my, cloud data set which, had about 1,500 images. Took. Just, about 2 to 2 to 3 hours to train although. As you saw in the UI the advanced, model depending on how many images you have can. Take up to four hours to Train because it's a much more complex. Type of model, so. It, really depends on how, many images are in your data set and. A complexity, of the. Variety of those images some. Tips to improve accuracy. As. I showed briefly, looking, at the evaluate, tab in the auto ml UI is, super useful I showed. The confusion, matrix there are a couple other metrics. There but. What I love about that confusion matrix in the UI is that you can actually click and see. What's being labeled correctly what's. Being labeled incorrectly so. I think that's the best way to improve accuracy on your model because. You want to remember that your, model is only going to be as good as a quality, of training data that you give it so. That that evaluate, tab can help you identify problems. With your training data so then you can go back look at the matrix and, see okay maybe a lot of my data was mislabeled, or maybe there just wasn't enough of this particular, label, category, in my data set and you.

Can Go back and improve it so. Once, here you read the next yeah let's go to the next one once, a model is built say, for identifying a particular brand, of toy who, owns that model Google or the customer who built it this, is a really important, question to get answered. You. Own your data. That's. An important point I'll say it again you own your data Google. Does not own the model we don't very. Important to mention that we. Don't use we do not use your data to retrain, our, models. For. Example Sarah. Showed the the cloud classification. That. Cloud, training, and the data and in the pictures themselves remain. Under. Sarah's, ownership, and we. There there's, a very hard. Break. Here that we don't use. That data or information to, Train anything, outside of it so, folks. That aren't worried, about intellectual, property or should I be you. Know how should i train my model, should I use my my, my special, sauce with with, my special, pictures you. Feel, that there is no danger of that getting out and we don't use that data to retrain, anything, else outside of. Your own models. Okay. I'll read the next one. We're. On the last one on this slide so so I want to classify on, a type, of image, or a brand, which. Isn't covered. By the vision API, what. Can I use to extend, a classification. That is already available through. Auto ml, / vision, API I think you touched on that a little bit yeah. Earlier. But I'll go, back into it so I mentioned, with the cloud example, so let's say that I just want to know is there a cloud in my image vision. API is great for that then I wanted to go one step further and say okay, what types of clouds are these maybe I'm using it to build a weather analysis application, then, I can train a custom, model using. Auto ml another. Example is let's say you've got a, product. That's, that's. A new product part, of your that your companies just launched and you want to identify all. The images that contain a picture of that product, since, its new the vision API will. Not be will not know what that specific product is since, that's specific to your company so then you, can build an auto ml model to train it to recognize all the images that contain that you could do the same with a brand, or logo of, a new. Product that you're launching you can build a custom, model that's able to identify, all, of the images that contain that brand in it. Awesome. I. Think. We these. Are all the questions that we have bubbled, up here. So. I think we're gonna move towards, our closing, then if we may so where, we. Have more content, coming down that, following. This session so we, invite you to stay tuned, the. Next session is around GG, sweeeeet security, and they're gonna be looking at the, enterprise. Mobility management, the.

And Also, the enterprise security protection, and much more so, thank you again for spending, last, couple minutes here with us we encourage you to go to cloud, Google, comm to, go deep, and wide on some of the subjects that we covered to you today and, thanks. For spending, some time with us take care thank, you. You.

2018-07-16 03:42

Show Video


This is so awesome! It's great that you are making something so complex accessible for more people. I can't wait to try AutoML Vision and see how it can improve outcomes.

Could you please share the GIST for the sample web app to query our model.

This is a great introduction to Google AutoML for App developers to use pre-trained model which are 5 models - speech, vision, natural language, video intelligence and translation which are main google cloud services - and explain tensorflow open source by google brain team which can achieve to build custom pre-trained model by Machine Learning developers. Thanks.

Other news