Welcome to this introductory course on AWS DeepLens. The world's first deep learning enabled video camera for developers. I'm Jyothi Nookula and I'm a senior product manager on the DeepLens team. Today, I'll show you how to get started building AWS DeepLens projects. We'll look at the device's hardware, the architecture, and some sample project templates. At the end, I'll demonstrate building a project using the object detection template from the AWS DeepLens console. Let's get started. At Amazon, we have been working on machine learning for 20 years. We know there's no easy way for you as a developer to learn machine learning, and test and prototype your ideas. So we decided to put deep learning into your hands literally. AWS DeepLens is a wireless enabled video camera and a development platform that's integrated with the AWS Cloud. It lets you use the latest artificial intelligence tools and technology to develop computer vision applications based on deep learning models. Before we start building projects, let's take a look at the actual device. What makes our AWS DeepLens stand apart is the on-board accelerator that is capable of delivering 100 gigaflops of compute, which means it can run 100 billion operations per second. So how do you take this device and actually build something? For that, you need to understand the general workflow for an AWS DeepLens deployment. To create and run an AWS DeepLens project, you typically use Amazon SageMaker, AWS lambda, and AWS green grass. You'd use Amazon SageMaker to train and validate a custom machine learning model or you could import a pre-trained model. AWS lambda functions in DeepLens perform three important operations. Pre-processing, capturing inference, and displaying output. Once a project is deployed to DeepLens, the model and the lambda function can run locally on the device. AWS DeepLens creates a computer vision application project that consists of the model and inference lambda function. AWS Green grass can deploy the project and a lambda run-time to your AWS DeepLens, as well as the software or configuration updates. This diagram illustrates how these services come together. First, when turned on, AWS DeepLens captures a video stream, it produces two output streams. The device stream, which is the video stream passed through without processing and the project stream, which contains the results of the model's processing video frames. From there, the inference lambda function receives unprocessed video frames and then passes those unprocessed frames to the project's deep learning model for processing. Finally, the inference lambda function receives the process frames back from the model and passes them on in the project stream. We don't really show it in the diagram, but AWS DeepLens is well integrated with other AWS services. For instance, projects deployed to AWS DeepLens are securely transferred using AWS Greengrass. The output of AWS DeepLens when connected to the Internet, can be sent back to the console via AWS IoT and Amazon Kinesis video streams. So now, we're ready to start building, right? The reality is, most of us don't have the skills to build a convolutional neural network model. That's why AWS DeepLens offers ready to deploy sample projects, which include a pre-trained convolutional neural network model and the corresponding inference function. AWS DeepLens offers seven sampled project templates ready-to-use, including templates such as object detection, hot dog/not hot dog, artistic style transfer, activity recognition, and face detection. With these sample projects, you can get started with machine learning in less than 10 minutes. These templated projects can be edited and extended. For instance, you could use the object detection project template to recognize when your dog is sitting on your couch and have the application send you an SMS to notify you of this event. Of course, you can also create, train, and deploy your own custom model to AWS DeepLens. The development platform supports a variety of deep learning frameworks including MXNet, TensorFlow, and Caffe. Now, let's take it for a spin. We'll deploy the object detection project. The project uses the single shot multi box detective framework to detect objects with a pre-trained resonate 15 network. The network has been trained on the Pascal VOC dataset and can recognize 20 different objects. The model takes the video stream from your AWS DeepLens as input and labels the objects that it identifies. It's a pre-trained optimized model that is ready to be deployed to your AWS DeepLens. After deploying it, you can review the objects your AWS DeepLens recognizes around you through the console. Let's see what it looks like in action. I'm starting here in the AWS management console. In the search bar, I'm going to type DeepLens. Now, we're looking at the project screen. If you don't see the project list view, click on the ''Hamburger menu'' on the left and select projects. Select "Create new project" on the top right, choose "User project template" and "Object detection", and scroll down the screen to choose next. I'm going to accept the default values, scroll to the bottom and select "Create". It can take a few moments for the project creation to fully complete. You can verify that the project was created successfully once the description field has a value. You may need to refresh your browser window if this value isn't populating. Now that the fields are populated, choose the radio button for the project and select deploy to device. On the target device screen, choose your device from the list and select review. Now, it's time to review the details of the deployment and select "Deploy". The process to deploy the model to your DeepLens can take a few minutes. You will know that it is completed when the blue banner changes to green. In the project output tab, copy the unique topic for your AWS DeepLens. Choose AWS IoT console and paste the unique topic, to subscribe to a topic. You can review the results of DeepLens detecting different objects in the room and printing the confidence code tool. Notice, as I change what the DeepLens can see, it will report that at the bottom of the screen. Now, let's change it to a bottle. Now, you have seen the power of DeepLens. I'm Jyothi Nookula with AWS. Thanks for watching.
