Revision as of 08:07, 3 December 2019

Getting Started

Host Environment

Ubuntu 18.04 (recommended) or 16.04

Force Recovery Mode

To enter force recovery mode, you can do:

1. Hold the Recovery key

2. Power on device

3. Wait for 5 seconds and you can release the Recovery key

Once it enters recovery mode successfully, the HDMI output should be disabled. Then, you have to connect a USB cable with TX2 device and PC. A new "nvidia apx" device will be detected on PC.

Flash Pre-built Image

First, make sure your TX2 device is already in Force Recover mode, and USB cable is connected.

Then, execute the TX2_flash.sh script which you can find it in the release folder.

$ sudo ./TX2_flash.sh

After script is done, the target device will boot into OS automatically.

Install SDK Components

Download the SDK Manager for Jetson TX2 series from JetPack website.

Note: You will need a nVidia developer account for access.

After download complete, install via dpkg.

$ sudo dpkg -i sdkmanager_0.9.14-4964_amd64.deb

Then, you're able to run SDK manager.

$ sdkmanager

Log in with your nVidia developer account, and you can see the STEP 01 page.

Demo

In this section, we setup and run demo applications on TX2 target device.

Export deepstream sdk root first.

$ export DS_SDK_ROOT="/opt/nvidia/deepstream/deepstream-4.0"

Deepstream Samples

There are 3 kinds of object detector demos in deepstream SDK.

To replace the video file, you can modify the corresponding config files. For example,

$ vim deepstream_app_config_yoloV3.txt
uri=file:///home/advrisc/Videos/2014.mp4

FasterRCNN

Setup:

$ cd $DS_SDK_ROOT/sources/objectDetector_FasterRCNN
$ wget --no-check-certificate  https://dl.dropboxusercontent.com/s/o6ii098bu51d139/faster_rcnn_models.tgz?dl=0  -O faster-rcnn.tgz
$ tar zxvf  faster-rcnn.tgz  -C .  --strip-components=1  --exclude=ZF_*
$ cp /usr/src/tensorrt/data/faster-rcnn/faster_rcnn_test_iplugin.prototxt .

$ make -C nvdsinfer_custom_impl_fasterRCNN

Run:

$ deepstream-app -c deepstream_app_config_fasterRCNN.txt

SSD

Setup:

$ cd $DS_SDK_ROOT/sources/objectDetector_SSD

$ cp /usr/src/tensorrt/data/ssd/ssd_coco_labels.txt .
$ pip install tensorflow-gpu

$ sudo apt-get install python-protobuf

$ wget http://download.tensorflow.org/models/object_detection/ssd_inception_v2_coco_2017_11_17.tar.gz
$ tar zxvf ssd_inception_v2_coco_2017_11_17.tar.gz
$ cd ssd_inception_v2_coco_2017_11_17
$ python3 /usr/lib/python3.6/dist-packages/uff/bin/convert_to_uff.py \
    frozen_inference_graph.pb  -O NMS  -p /usr/src/tensorrt/samples/sampleUffSSD/config.py  -o sample_ssd_relu6.uff
$ cp sample_ssd_relu6.uff ../

$ cd ..
$ make -C nvdsinfer_custom_impl_ssd

Run:

$ deepstream-app -c deepstream_app_config_ssd.txt

Yolo

Setup:

$ cd $DS_SDK_ROOT/sources/objectDetector_Yolo
$ ./prebuild.sh
$ export CUDA_VER=10.0
$ make -C nvdsinfer_custom_impl_Yolo

Run:

$ deepstream-app -c deepstream_app_config_yoloV3.txt
-OR-
$ deepstream-app -c deepstream_app_config_yoloV3_tiny.txt

Deepstream Reference Apps

In this repository, it provides some reference applications for video analytics tasks using TensorRT and DeepSTream SDK 4.0.

$ cd $DS_SDK_ROOT/sources/apps/sample_apps/
$ git clone https://github.com/NVIDIA-AI-IOT/deepstream_reference_apps.git
$ cd deepstream_reference_apps

back-to-back-detectors & anomaly

These two applications only support elementary h264 stream, not mp4 video file.

runtime_source_add_delete

Setup:

$ cd runtime_source_add_delete
$ make

Run:

$ ./deepstream-test-rt-src-add-del <uri>
$ ./deepstream-test-rt-src-add-del file://$DS_SDK_ROOT/samples/streams/sample_1080p_h265.mp4
$ ./deepstream-test-rt-src-add-del rtsp://127.0.0.1/video