In order to recognize small objects, we need to crop the image， so that nvinfer can get objects
Gst-nvinfer can crop multiple objects images when it works as SGIE. Please refer to Gst-nvinfer — DeepStream 5.1 Release documentation and the sample code of deepstream-test2 (C/C++ Sample Apps Source Details — DeepStream 5.1 Release documentation).
In order to identify small targets，I want to divide a 1920 * 1080 image into four images，and then, after “nvstreammux” and “nvinfer”. what should we do in deepstream
Hi, I want to do exactly the same. I could split the video image, and apply four videocrops, but it really seems overkill and underperformant.
This definitely should be supported in deepstream. Have you found any way to do it?
Thanks a lot.