This code demonstrates Contour Detection using the Nvidia Performance Primitives (NPP+) library. Contour detection involves identifying the boundaries of objects within an image, which is a key task in computer vision and image processing.
- Image Contour Detection: Detecting the edges and boundaries of objects within images to analyze shapes, objects, and regions of interest.
This example is optimized for the following Nvidia GPU architectures:
- Linux
- Windows
- x86_64
To begin using NPP+, download the latest version of the library and install it on your Nvidia GPU-equipped machine.
This example uses the NPP+ library, which provides optimized functions for image processing tasks on Nvidia GPUs.
This project includes the following main components:
- Find Contour Sample: Demonstrates how to efficiently detect contours in images using NPP+ functions.
Before building and running the example, ensure you have the following:
- A Linux or Windows system with a supported Nvidia GPU.
- The latest NVIDIA drivers installed.
- Install the CUDA 12.0 toolkit and above.
$ mkdir build
$ cd build
$ cmake .. -DNPP_PLUS_PATH=/usr/lib/x86_64-linux-gnu
$ make
$ mkdir build
$ cd build
$ cmake -DCMAKE_GENERATOR_PLATFORM=x64 ..
$ Open findContour.sln project in Visual Studio 15 2017 and build
./findContour -h
Usage: ./findContour
Parameters:
number-of-images : Use 2 number of images
Example:
$ ./findContour
./findContour
NPP Library Version 0.9.0 CUDA Driver Version: 12.7 CUDA Runtime Version: 12.6
Input file load succeeded.
CircuitBoard_CompressedMarkerLabelsUF_8Way_2048x1024_32u succeeded, compressed label count is 249.
nID 1 BB 190 0 539 299
ID 1 Cnt 1 505 285
ID 1 Cnt 2 539 275
ID 1 Cnt 3 533 263
ID 1 Cnt 4 537 269
ID 1 Cnt 5 513 265
ID 1 Cnt 6 512 265
ID 1 Cnt 7 511 26
Input Image
Input Image - CircuitBoard_2048x1024_8u.raw - image name has image info such as width=2048, height=1024 and data type=8bit unsigned.
Output images at different stages of Contour – Figure 1 shows (CircuitBoard_LabelMarkersUF_8Way_2048x1024_32u.raw) the results from the nppiLabelMarkersUF_8u32u_C1R_Ctx() API and which generate the 1 channel 8-bit to 32-bit unsigned integer label markers image.
Image info – 8 way connectivity width=2048, height=1024 and datatype=32bit unsigned
Figure 2 shows (CircuitBoard_CompressedMarkerLabelsUF_8Way_2048x1024_32u.raw) the results from the nppiCompressMarkerLabelsUF_32u_C1IR() API and typical operation 1 channel 32-bit unsigned integer in place connected region marker label renumbering for output from nppiLabelMarkersUF functions only with numbering sparseness elimination.
Image info – 8 way connectivity width=2048, height=1024 and datatype=32bit unsigned
Figure 3 shows (CircuitBoard_Contours_8Way_2048x1024_8u.raw) the results from the nppiCompressedMarkerLabelsUFInfo_32u_C1R_Ctx() API and typical operation 1 channel 32-bit unsigned integer connected region marker label renumbered from a previous call to nppiCompressMarkerLabelsUF or nppiCmpressMarkerLabelsUFBatch functions to eliminate label ID sparseness.
Image info – 8 way connectivity width=2048, height=1024 and datatype=8bit unsigned
Figure 4 shows (CircuitBoard_ContoursReconstructed_8Way_2048x1024_8u.raw) the results from the nppiCompressedMarkerLabelsUFContoursOutputGeometryLists_C1R() API and typical operation includes 1 channel connected region contours image to output contours geometry list in host memory. Note that ALL input and output data for the function MUST be in host memory. Also nFirstContourID and nLastContourID allow only a portion of the contour geometry lists in the image to be output. Note that the geometry list for each contour will begin at pContoursGeometryListsHost[pContoursPixelStartingOffsetHost[nContourID] * sizeof(NppiContourPixelGeometryInfo). Also note that the ordered contour geometry list is contained in the oContourOrderedGeometryLocation object within the NppiContourPixelGeometryInfo object for each contour pixel and that this location information is the only valid information in the object relevant to that ordered pixel
Image info – 8 way connectivity width=2048, height=1024 and datatype=8bit unsigned
