Jetson Naon DeepStream SDK Getting Started

DeepStream SDK 基於 GStreamer library. 透過plug-in的方式使用GPU來處理串流資料. 例如Encode / Decode H.264串流, 物件辨識及目標追蹤, On Screen Display標示物件等.

1.Install DeepStream SDK - 4.0.2

https://docs.nvidia.com/metropolis/deepstream/dev-guide/

最容易的方式是Method 3 以Debian package安裝, 相依的套件會全部自動安裝

$ sudo apt-get install ./deepstream-4.0_4.0.2-1_arm64.deb

安裝完成後 DeepStream SDK 在 /opt/nvidia/deepstream/deepstream-4.0

2.DeepStream SDK的測試範例

$ cd /opt/nvidia/deepstream/deepstream-4.0/sources/apps/sample_apps/deepstream-test1

編輯檔案 dstest1_pgie_config.txt, 修改network-mode (0 : FP32, 1 : Int8, 2 : FP16)

因為 Jetson Naon 不支援 Int8 inference

#network-mode=1
network-mode=2

$ deepstream-test1-app /opt/nvidia/deepstream/deepstream-4.0/samples/streams/sample_720p.h264

執行後要等一陣子產生model engine之後才以此處理並輸出影像串流

執行完成後model engine file會儲存起來,這時再度編輯dstest1_pgie_config.txt, 修改

model-engine-file=../../../../samples/models/Primary_Detector/resnet10.caffemodel_b1_fp16.engine

再次執行deepstream-test1-app就會直接使用之前產生的 .engine file

若出現以下錯誤

No EGL Display 

nvbufsurftransform: Could not get EGL display connection

編輯 ~/.bashrc 在檔案最後加上

unset DISPLAY

export DISPLAY=:0

執行 source ~/.bashrc

修改 /etc/ssh/sshd_config

+   #X11Forwarding no

-    X11Forwarding yes

執行 xrandr 確認 X11有跑起來

3.DeepStream Application

由以上的測試範例可知DeepStream Application是以c code來寫程式組合GStreamer串流, Inference則以 .txt 作為設定檔, 設定串流的source / sink, inference的各項參數等等

4.DeepStream Yolo

Yolo物件辨識的範例

$ cd /opt/nvidia/deepstream/deepstream-4.0/sources/objectDetector_Yolo
$ ./prebuild.sh
$ cd nvdsinfer_custom_impl_Yolo/
$ export CUDA_VER=10.0
$ make
$ cd ../

採用YoloV3 tiny

編輯 config_infer_primary_yoloV3_tiny.txt, 修改network-mode (0 : FP32, 1 : Int8, 2 : FP16)

#network-mode=1
network-mode=2

model-engine-file=model_b1_fp16.engine

$ deepstream-app -c deepstream_app_config_yoloV3_tiny.txt

Using winsys: x11 

Creating LL OSD context new

Deserialize yoloLayerV3 plugin: yolo_17

Deserialize yoloLayerV3 plugin: yolo_24

Runtime commands:

h: Print this help

q: Quit

p: Pause

r: Resume

NOTE: To expand a source in the 2D tiled display and view object details, left-click on the source.

      To go back to the tiled display, right-click anywhere on the window.

**PERF: FPS 0 (Avg)

**PERF: 0.00 (0.00)

** INFO: <bus_callback:189>: Pipeline ready

Opening in BLOCKING MODE 

NvMMLiteOpen : Block : BlockType = 261 

NVMEDIA: Reading vendor.tegra.display-size : status: 6 

NvMMLiteBlockCreate : Block : BlockType = 261 

** INFO: <bus_callback:175>: Pipeline running

Creating LL OSD context new

**PERF: 29.62 (29.62)

**PERF: 29.80 (29.71)

**PERF: 29.94 (29.79)

**PERF: 29.78 (29.79)

**PERF: 29.92 (29.81)

**PERF: 29.61 (29.78)

**PERF: 29.69 (29.77)

**PERF: 29.66 (29.75)

**PERF: 29.80 (29.76)

** INFO: <bus_callback:212>: Received EOS. Exiting ...

Quitting

App run successful