we have a gstreamer pipeline which consists of nvinfer and nvtracker. Because number of objects are alot I only wanna select objects which are near to center of image. my camera resolution is full-HD and I have trained detection network with full-HD size. I want nvinfer input size be 1920*1080 (because it’s my training size and has the most accuracy) but detection occurs only in a HD size of image center which is in bellow figure or if not possible, detection occurs on full-HD but only objects in image center HD pass to nvtracker.
Can I do that in nvinfer?
if yes which function?