Not able to access raw tensor output as metadata

[theelitepro0224]

Thanks for prompt response.

@fanzh

the format is always nv12 after streamdemux.

So has this changed from rgb to NV12 in DS7.1? If not then why this code was working with 7.0 and not with 7.1?

What are the changes required in the code achieve the same.

@theelitepro0224

cv::Mat frame = cv::Mat(height, width, CV_8UC4, frame_data, pitch);

[fanzh]

please refer to this faq Work on opencv nv12 mat. please try the following code for creating a Mat for NV12.

Mat nv12_mat = Mat(height*3/2, width, CV_8UC1, surface->surfaceList[frame_meta->batch_id].mappedAddr.addr[0],

[theelitepro0224]

I have used the code from provided link. The provided code needs to be updated for opencv 4.10 and DS7.1. It is bit old. However, I have been able to copy to Mat using for loop and memcpy and then converting Mat to Surface using SurfaceTransform API.

I am facing a problem now that my custom libnvdsgst_infer.so is not able to execute my custom libnvds_infer.so. Using ldd command I have confirmed that libnvdsgst_infer.so is looking at the path where libnvds_infer.so is residing. If I use relative path in the Makefile then I face this issue while if I use absolute path then I don’t see it.

LIB:=libnvds_infer.so

NVDS_VERSION:=7.1

#LIB_INSTALL_DIR?=/opt/nvidia/deepstream/deepstream-$(NVDS_VERSION)/lib/
LIB_INSTALL_DIR?=../../../lib/

LIB:=libnvdsgst_infer.so

NVDS_VERSION:=7.1

CFLAGS+= -fPIC -std=c++11 -DDS_VERSION=\"7.1.0\" \
	 -I /usr/local/cuda-$(CUDA_VER)/include \
	 -I ../../includes \
	 -I ../gst-nvdspreprocess/include \
	 -I ../../libs/nvdsinfer -DNDEBUG

#GST_INSTALL_DIR?=/opt/nvidia/deepstream/deepstream-$(NVDS_VERSION)/lib/gst-plugins/
GST_INSTALL_DIR?=../../../lib/gst-plugins/
#LIB_INSTALL_DIR?=/opt/nvidia/deepstream/deepstream-$(NVDS_VERSION)/lib/
LIB_INSTALL_DIR?=../../../lib/

Here LIB_INSTALL_DIR need absolute path. Please resolve.

[fanzh]

seems the orginal issue was fixed. could you open a new topc for the new issue? Thanks!

[theelitepro0224]

Thanks for your continued support. I have raised a new issue Custom libnvdsgst_infer not linking custom libnvds_infer during runtime

Request you to please take a look at it.