一起学习Deepstream视频分析课程

NVIDIA DLI发布了一个最新的免费课程：

让我们一起来学习吧。

首先打开网址：https://courses.nvidia.com/courses/course-v1:DLI+C-IV-02+V1/about

看一下这个课程的学习内容：

You'll learn how to:

-Set up your Jetson Nano and (optional) camera

-Build end-to-end DeepStream pipelines to convert raw video input into insightful annotated video output

-Configure multiple video streams simultaneously

注意，虽然课程是基于Jetson NANO的，其实Xavier和Jetson TX2都是支持的哟！

好了，报名这个课程，点击进入吧

进入后，我们看到课程要求：

Hardware：

Jetson Nano Developer Kit

Computer with Internet Access and SD card port

A microSD Memory Card (32GB UHS-I minimum)

Compatible 5V 4A Power Supply with 2.1mm DC barrel connector

2-pin jumper

USB cable (Micro-B to Type-A)

A computer with an Internet connection and:

-The ability to flash your microSD card

-Administrative rights and the ability to install a compatible media player software such as VLC Player

OPTIONAL: Compatible camera such as a Logitech C270 Webcam

OPTIONAL: Wired Internet connection to the Jetson Nano (Ethernet port)

简单地说，就是你需要有一套Jetson NANO开发套件，上面有TF卡，至少32GB，刷好系统，有5V4A电源，有跳线帽；准备一台联网的电脑，电脑上装一个媒体播放器，比如VLC；可以准备一个USB摄像头哦，比如罗技C270.（实际上跟我们购买Jetson NANO套件的套餐，会包含32G刷好系统的TF卡和5V4A电源。升级版本的Jetson NANO是自带跳线帽的,

不过我们建议用户用一个全新的32GB TF卡，根据这个课程学会自己刷机安装系统）。

好了，点击课程进入下一个画面，我们会看到整个课程目录：

大家可以按照这个课程一步一步看。我们来到最后的问答部分：

1. Which of the following statements are true about "bounding boxes" in the context of object detection? (Check all that apply)。

- Bounding boxes are used to show target locations.

-A bounding box is a rectangular box determined with x and y coordinates of the axis.

-A bounding box is used to shrink the size of the overall image.

-In a DeepStream pipeline, the Gst-nvdsosd plugin is used to draw bounding boxes, text, and region-of-interest (RoI) polygons.

2.What is the Gst-nvinfer plugin used for? (Check all that apply)

-Performs transforms (format conversion and scaling), on the input frame based on network requirements, and passes the transformed data to the low-level library

-Performs inferencing on input data using NVIDIA TensorRT

-Sends UDP packets to the network.

-Tracks object between frames.

3.What feature describes a "hardware-accelerated" plugin?

-Interacts with hardware to deliver maximum performance such as GPU, DLA, PVA

-Improves High level software interactions such as Python or Java

-Executes on embedded processors

4.Looking at the config file on your Jetson Nano at “/home/dlinano/deepstream_sdk_v4.0.2_jetson/sources/apps/dli_apps/deepstream-test1-rtsp_out/dstest1_pgie_config.txt”: What can we understand about the model we are using for object detection and counting? (Hint: check out the "model-file", "num-detected-classes", and "output-blob-names" keys)

-ResNet-10, Number of Classes 4, Output Layer: conv2d_bbox

-AlexNet, Number of Classes 100, Output Layer: conv2d_bbox

-ResNet-10, Number of Classes 100, Output Layer: conv2d_bbox

-AlexNet, Number of Classes 1000, Output Layer: conv2d_bbox

5.Looking at the “C” file at “/home/dlinano/deepstream_sdk_v4.0.2_jetson/sources/apps/dli_apps/deepstream-test1-rtsp_out/deepstream_test1_app.c”: In line 67, we use the “NvDsBatchMeta”, metadata structure. Why is it needed? (Check all that apply)

-It is not actually used.

-We need a metadata structure to hold frames, object, classifier, and label data.

-We need to access the metadata in this structure to determine how many objects are in each frame and display them

-It is a piece of legacy code carried forward from a previous version of DeepStream.

6.What type(s) of network(s) are supported by Gst-nvinfer?

-Multi-class object detection only

-Multi-label classification only

-Multi-class object detection, multi-label classification, and segmentation

7.What is Gst-nvstreammux used for? (Check all that apply)

-It forms a batch of frames from one or multiple input sources

-It collects an average of ( batch-size/num-source ) frames per batch from each source

-It runs inference

-It tracks objects between frames

8.How should we determine the batch size for multiple stream inputs?

-It should be equal to or proportional to the number of input streams

-Batch size is inversely proportional to number of streams

-Batch size can be 1 even for multiple stream inputs

9.Which of the following plugin lists are included in both Notebook #2: "Multiple Networks Application" and Notebook #3: "Multiple Stream Inputs"?

-Gst-nvinfer (for deep learning inference), Gst-nvstreammux (for batching video streams), Gst-nvtracker (tracks object between frames)

-Gst-nvinfer (for deep learning inference), Gst-nvdsosd (for drawing bounding boxes), Gst-nvvideoconvert (for video format conversion)

-Gst-nvinfer (for deep learning inference), Gst-nvmultistreamtiler (composites a 2D tile from batched buffers), Gst-nvtracker (tracks object between frames)

-Gst-nvvideoconvert (for video format conversion), Gst-nvstreammux (for batching video streams), Gst-nvmultistreamtiler (composites a 2D tile from batched buffers)

10.What are the DeepStream supported object detection networks? (Check all that apply)

-ResNet-10

-YOLO

-SSD -Faster-RCNN

11.What can be understood by looking at the config file “/home/dlinano/deepstream_sdk_v4.0.2_jetson/sources/apps/dli_apps/deepstream-test3-mp4_out-yolo/dstest3_pgie_config_yolov3_tiny.txt”?

-Neural Network Model : YOLO-V3, Number of Classes: 4, Network Mode= FP32 (Floating Point Computations), Input to model in format: BGR

-Neural Network Model : Tiny-YOLO-V3, Number of Classes: 80, Network Mode= FP32 (Floating Point Computations), Input to model in format: RGB

-Neural Network Model : Tiny-YOLO-V3, Number of Classes: 80, Network Mode= INT8 (Floating Point Computations), Input to model in format: BGR

12.Which of the following DeepStream plugins are part of the Primary Detector -> Object Tracker -> Secondary Classifier(s) sequence used in the pipeline for Multiple Network Applications in Notebook #2? (Check all that apply):

-Gst-nvinfer

-Gst-nvvideoconvert

-Gst-nvmultistreamtiler

-Gst-nvtracker

13.What kind of video container (file type) can a DeepStream video output be saved to? (Check all that apply): -mp4

-avi

-anything GStreamer supports

14.Which of the following statements are true about DeepStream SDK? (Check all that apply)

-DeepStream SDK is based on the GStreamer framework

-DeepStream SDK is not designed to optimize performance

-DeepStream SDK is supported on systems that contain NVIDIA Jetson modules, and NVIDIA dGPU adapters.

-DeepStream SDK has a plugin interface for TensorRT for inferencing deep learning networks.

15.Which of the following are possible use cases of DeepStream? (Check all that apply)

-Intelligent Video Analytics

-AI-based video and image understanding

-Multi-sensor processing -Cloud-based offline processing

做完所有的题目点击一下步，最后到Survey这里，“下一步”显示为灰色，表明已经完成了所有的课程。然后点击最上面的“进度”

就能看到完整的进度报告——以及可以打印证书！

好了，点击申请证书，见证奇迹的时刻吧！