Exercise - Modify the DeepStream sample applications

• 5 minutes

Now that you've run a DeepStream sample application, you can modify the samples to get different behavior.

1. Let's start by viewing the sample configuration's structure in a text editor. We'll make a copy of the sample configuration used in the previous section and look at some settings we can change. Run these commands:

   cd /opt/nvidia/deepstream/deepstream-6.0/samples/configs/deepstream-app
   sudo cp source4_1080p_dec_infer-resnet_tracker_sgie_tiled_display_int8.txt source4_1080p_dec_infer-resnet_tracker_sgie_tiled_display_int8_modified.txt
   sudo vi source4_1080p_dec_infer-resnet_tracker_sgie_tiled_display_int8_modified.txt
   

   Note

   Here are some tips for using vi when you edit the source4_1080p_dec_infer-resnet_tracker_sgie_tiled_display_int8_modified.txt file:

   • Press the i key to put the editor into Insert mode. Then you'll be able to make changes.
   • Press Esc to go exit Insert mode and return to Normal mode.
   • To save and quit, enter :x, and then press Enter.
   • Save the file, enter :w, and then press Enter.
   • To close vi, enter :quit, and then press Enter.

2. Note the various configuration sections and how they contribute to the overall application. These sections are denoted with brackets ([]). For example, [application], [tiled-display], [source0], and [sink0]. These sections are explained in detail in the Configuration Groups section of the DeepStream SDK documentation.

   For example, to change the input to use an RTSP video stream instead of a local video file, change [tiled-display] and [source0]:

   [tiled-display]
   enable=1
   rows=2
   columns=2
   width=1280
   height=720
   gpu-id=0
   #(0): nvbuf-mem-default - Default memory allocated, specific to particular platform
   #(1): nvbuf-mem-cuda-pinned - Allocate Pinned/Host cuda memory, applicable for Tesla
   #(2): nvbuf-mem-cuda-device - Allocate Device cuda memory, applicable for Tesla
   #(3): nvbuf-mem-cuda-unified - Allocate Unified cuda memory, applicable for Tesla
   #(4): nvbuf-mem-surface-array - Allocate Surface Array memory, applicable for Jetson
   nvbuf-memory-type=0
   
   [source0]
   enable=1
   #Type - 1=CameraV4L2 2=URI 3=MultiURI 4=RTSP
   type=3
   uri=file://../../streams/sample_1080p_h264.mp4
   num-sources=4
   #drop-frame-interval=2
   gpu-id=0
   # (0): memtype_device   - Memory type Device
   # (1): memtype_pinned   - Memory type Host Pinned
   # (2): memtype_unified  - Memory type Unified
   cudadec-memtype=0
   

   Then make the following changes. (Note the changes to the rows, columns, type, uri, and num-sources fields.)

   [tiled-display]
   enable=1
   rows=1
   columns=1
   width=1280
   height=720
   gpu-id=0
   #(0): nvbuf-mem-default - Default memory allocated, specific to particular platform
   #(1): nvbuf-mem-cuda-pinned - Allocate Pinned/Host cuda memory, applicable for Tesla
   #(2): nvbuf-mem-cuda-device - Allocate Device cuda memory, applicable for Tesla
   #(3): nvbuf-mem-cuda-unified - Allocate Unified cuda memory, applicable for Tesla
   #(4): nvbuf-mem-surface-array - Allocate Surface Array memory, applicable for Jetson
   nvbuf-memory-type=0
   
   [source0]
   enable=1
   #Type - 1=CameraV4L2 2=URI 3=MultiURI 4=RTSP
   type=4
   uri=rtsp://wowzaec2demo.streamlock.net/vod/mp4:BigBuckBunny_115k.mov
   num-sources=1
   #drop-frame-interval=2
   gpu-id=0
   # (0): memtype_device   - Memory type Device
   # (1): memtype_pinned   - Memory type Host Pinned
   # (2): memtype_unified  - Memory type Unified
   cudadec-memtype=0
   
   

3. Run the changed configuration by using these commands:

   cd /opt/nvidia/deepstream/deepstream-6.0/samples/configs/deepstream-app
   deepstream-app -c source4_1080p_dec_infer-resnet_tracker_sgie_tiled_display_int8_modified.txt
   

Try this

Look at the various sample configurations included in the DeepStream SDK. How might you change these samples to create a smart-home security system? Assume you have multiple RTSP cameras that serve feeds over unique RTSP endpoints. Could you create a DeepStream application that uses those live video streams as input and apply inference to detect people and vehicles?

Next steps

To finish this module and earn your trophy, complete a final knowledge check.