Intelligent Pipeline Generator Demo Beta Release

Hey everyone, I am Aayush and in this video I'm going to show the latest features of uh the intelligent pipeline generator with this new release. We have custom plug-in generation and support for Nvidia Jets and devices. So let's get started. First of all, uh once you have your application ready, you can log into application. Then we'll create a new project. And this project is for person segmentation which is basically a segmentation model uh for people and this model is not originally supported by nstreamer. So we'll create a custom plug-in. So first of all we just need to choose a model file and we'll open the terminal and then just navigate to the project directory. And once we are in the project directory, we just need to run our project. Yes, there it is. So that's the segmentation. It the model segments me as in a red color. Now we'll generate the architecture. And so this is the custom plug-in generation tab where we can generate the custom plug-in. So once architecture is generated, we can check if custom decoder is required or not. And now it shows image segmentation task for the model. Now we can click on generate plug-in and then this coder will automatically analyze the requirements of our project and start working on it. So uh it also lists what uh plugins are already installed on our system for tensor decoder. Basically it is a subplugin for tensor decoder and it is still analyzing it takes some time and then it starts writing the code. So this is the plug-in code. We have the options to compile it or edit it. We have a built-in editor that allows us to modify the code and it has all features like find like you would normally find in any normal code editor. So either you can edit it or compile it. It'll ask for password to place the compile so file into the directory where it's supposed to be. And now you can see that a new plug-in named custom personse IPG is shown as an entry here. Now we can just go ahead and generate pipeline. The pipeline generator agent already knows that there is a new plug-in. So it sets the mode automatically. Now just click on run pipeline and it works. It segmented me. The color is randomized. So it'll give a random color every time. Let us create a new project for YOLO X. This model is also not supported by NN streamer and it's a different model. It's for object detection. Again we submit the model. We change to the project directory and run the uh application. And yes it works. So now I can click on submit trace data. It will submit all the relevant files it has traced that I used to run the model. Then click on generate architecture and wait for it to load. Basically this uh custom plug-in section identifies and presents like what the model is, what is the task of the model and what the plug-in is being generated for. Now it has identified that it's bounding boxes. So you can just click on generate plug-in and it again starts to analyze and once it's done it starts to code that plug-in in C. So these codes are pretty complex and it's AI based so it may not work every time. That's why we have a code editor in case we want to edit uh the code ourselves manually and it acts as a good starting point if you want to code the plug-in uh in case it doesn't work. So as you can see these uh there are too many lines of code for the decoding of a single model output and once it's done it will create and install another plug-in. So the plugins that it creates they start with custom and ends with IPG and in the middle they have the same name as the project name. So that is what uh differentiates the plugins that are generated by this application and I think it's about to get completed. Okay, that was really long. It's completed. Now we can compile the plug-in. Enter the password and it's installed. Now we can see yes it is installed at this location. Now we can just generate another pipeline and again it'll know that we have a plug-in which is of this name. Just click on run pipeline and it works. It's able to draw the bounding boxes. Now we have uh switched to a Nvidia Jetson device. Let us open the application on our Jetson device and I'll demonstrate how we can use to build applications in deep stream that runs on Nvidia Jets and devices. Just name your model. This example we're using RTDR uh detection model. It's again for object detection. Now this is much simpler. We just need to select the file that's the Python file and our model file and then we need to just specify for example if we using a video here we just need to specify the path of this video file. We can enter other parameters here as well. Just enter the password after clicking generate deep stream app and it'll generate the app and then just run deepream app. Yeah, here it is. So, as you can see, uh it's able to draw these detections. So um that is it. Uh thanks for watching.