L4T Multimedia API Reference27.1 Release |
The camera with CAFFE object identification sample demonstrates how to capture the image from the camera, convert the data through NvVideoConverter without memory copy, feed the memory pointer to OpenCV, and run CAFFE model zoo for the object identication algorithm.
This sample adds zoo object identification into part of the consumer thread. This is achieved via the opencv_img_processing()
function invoked by the capture plane callback. The code is located under opencv_consumer_lib
to integrate deep learning function as a sample. Frame rate is not part of the key consideration.
CAFFE object identication uses OpenCV Mat as the input and output structure. For more information, see:
Before running the sample, you must have the following:
The struct context_t
global structure manages all of the resources in the application.
The NvVideoConverter class packages all video converting related elements and functions. The sample uses the following key members:
NvVideoConverter | Description |
---|---|
output_plane | Specifies the output plane. |
capture_plane | Specifies the capture plane. |
waitForIdle | - |
setOutputPlaneFormat | Sets the output plane format. |
setCapturePlaneFormat | Sets the capture plane format. |