NVIDIA VPI GStreamer Plug-in - Examples - Separable Convolution

GstVPI
GstVPI Basics
Getting Started
Evaluating GstVPI Getting the Code Building the Plugin
Examples
Gaussian Filter Undistort Harris Keypoints Detector KLT Tracker Perspective Warp Overlay Videoconvert Videoscale Convolution Separable Convolution
Performance
Nano TX1 TX2 Xavier
Contact Us

Introduction to Separable Convolution

The vpiseparableconvolution element applies the Separable Convolution algorithm from the NVIDIA VPI library to a video stream. In order to use this algorithm, you must provide the kernel row and kernel column to be applied.

Element properties

• backend
  

Backend to use to execute VPI algorithms. Available options:

cpu: CPU backend

cuda: CUDA backend

pva: PVA backend (Xavier only)

vic: VIC backend

Flags: readable, writable
Default: "cuda"

• kernel-row
  

Horizontal convolution kernel to be applied with a minimum size of 1 and maximum of 11. Usage example:

<1.0, 0.0, -1.0>

Type: GstValueArray of GValues
Flags: readable, writable
Default: <1>

• kernel-col
  

Vertical convolution kernel to be applied with a minimum size of 1 and maximum of 11. Usage example:

<1.0, 0.0, -1.0>

Type: GstValueArray of GValues
Flags: readable, writable
Default: <1>

• boundary
  

How pixel values outside of the image domain should be treated. Available options:

zero: All pixels outside the image are considered 0.

clamp: Border pixels are repeated indefinitely.

Flags: readable, writable
Default: "zero"

NVIDIA VPI GStreamer Plug-in Example for Separable Convolution

The following example pipe will help you apply separable convolution to the input video stream from your camera. You may modify the properties values according to the information above.

For a separable convolution given by a 7x7 Sobel filter, and a boundary condition of zero you may use:

gst-launch-1.0 nvarguscamerasrc ! nvvidconv ! 'video/x-raw,format=GRAY8' ! vpiupload ! vpiseparableconvolution kernel-row="<-1.0, -5.0, -6.0, 0.0, 6.0, 5.0, 1.0>" kernel-col="<0.015625, 0.09374, 0.234375, 0.3125, 0.09374, 0.015625>" ! vpidownload ! nvvidconv ! 'video/x-raw(memory:NVMM),format=I420' ! nvoverlaysink

Another example pipeline for a video test source could be the following:

gst-launch-1.0 videotestsrc pattern=spokes ! 'video/x-raw,format=GRAY8' ! vpiupload ! vpiseparableconvolution kernel-row="<-1.0, -5.0, -6.0, 0.0, 6.0, 5.0, 1.0>" kernel-col="<0.015625, 0.09374, 0.234375, 0.3125, 0.09374, 0.015625>" ! vpidownload ! nvvidconv ! "video/x-raw(memory:NVMM),format=I420" ! nvoverlaysink

Alternative display pipeline.

gst-launch-1.0 videotestsrc pattern=spokes ! 'video/x-raw,format=GRAY8' ! vpiupload ! vpiseparableconvolution kernel-row="<-1.0, -5.0, -6.0, 0.0, 6.0, 5.0, 1.0>" kernel-col="<0.015625, 0.09374, 0.234375, 0.3125, 0.09374, 0.015625>" ! vpidownload ! nvvidconv ! "video/x-raw(memory:NVMM),format=I420" ! nvvidconv ! autovideosink