How to change nvbuffer to cpu-buffer when nvarguscamerasrc is working in bufapi-version mode？

[blstcvLiu]

The pipeline structure : nvarguscamesrc(cap:NVMM/NV12,25fps,4000*3000)->nvvidconvert->appsink.
I got the nvmm-buffer in the appsink’s “new-sample”-callback,then I can change nvbuffer to cpu-buffer when nvarguscamerasrc was working without bufapi-version by using the following API:

Now I want to add nvinfer to the pipeline,so the nvarguscamesrc need to set “bufapi-version” from false to true,
but over this operated, the map.size was change to 64 and the api -“ExtractFdFromNvBuffer” return error.

How to change nvbuffer to cpu-buffer when nvarguscamerasrc is working in bufapi-version mode？

[DaneLLL]

Hi,
For setting bufapi-version=true, please use NvBufSurface APIs. You may refer to

/opt/nvidia/deepstream/deepstream-6.0/sources/gst-plugins/gst-dsexample/gstdsexample.cpp

The demo code is in get_converted_mat(). Please also check APIs in

/opt/nvidia/deepstream/deepstream-6.0/sources/includes/nvbufsurface.h
/opt/nvidia/deepstream/deepstream-6.0/sources/includes/nvbufsurftransform.h

[blstcvLiu]

Hello,DaneLLL
I had got Segmentation fault when dealing with a copy of a nvmemory map，whether NV12 data requires additional operations?
The test code is shown below(Segmentation falut position :memcpy((char*)snap_yuv_img.data,dst_surface->surfaceList[0].mappedAddr.addr[0],400030003/2);):

GstBuffer *gstbuffer = gst_sample_get_buffer(sample);
if(gstbuffer)
{
GstMapInfo map;
if (gst_buffer_map(gstbuffer, &map, GST_MAP_READ))
{
if(0 < need_snap_counter && 0 == is_snap_process && 0 == storage_spacefull_flag)
{
NvBufSurface *surface = NULL;
NvBufSurfTransformRect src_rect, dst_rect;
surface = (NvBufSurface *) map.data;
int batch_size= surface->batchSize;
src_rect.top = 0;
src_rect.left = 0;
src_rect.width = (guint) surface->surfaceList[0].width;
src_rect.height= (guint) surface->surfaceList[0].height;
dst_rect.top = 0;
dst_rect.left = 0;
dst_rect.width = 4000;
dst_rect.height= 3000;

		NvBufSurfTransformParams nvbufsurface_params;
		nvbufsurface_params.src_rect = &src_rect;
		nvbufsurface_params.dst_rect = &dst_rect;
		nvbufsurface_params.transform_flag = NVBUFSURF_TRANSFORM_NORMALIZE; //NVBUFSURF_TRANSFORM_CROP_SRC | NVBUFSURF_TRANSFORM_CROP_DST;
		nvbufsurface_params.transform_filter = NvBufSurfTransformInter_Default;

		NvBufSurface *dst_surface = NULL;
		NvBufSurfaceCreateParams nvbufsurface_create_params;

		nvbufsurface_create_params.gpuId  = surface->gpuId;
		nvbufsurface_create_params.width  = (gint) surface->surfaceList[0].width;
		nvbufsurface_create_params.height = (gint) surface->surfaceList[0].height;
		nvbufsurface_create_params.size = 0;
		nvbufsurface_create_params.colorFormat = NVBUF_COLOR_FORMAT_NV12;
		nvbufsurface_create_params.layout = NVBUF_LAYOUT_PITCH;
		nvbufsurface_create_params.memType = NVBUF_MEM_DEFAULT;
		
		cudaSetDevice (surface->gpuId);
		cudaStream_t cuda_stream;
		cudaStreamCreate (&cuda_stream);
		int create_result = NvBufSurfaceCreate(&dst_surface,batch_size,&nvbufsurface_create_params);
		
		NvBufSurfTransformConfigParams transform_config_params;
		NvBufSurfTransform_Error err;
		
		transform_config_params.compute_mode = NvBufSurfTransformCompute_Default;
		transform_config_params.gpu_id = surface->gpuId;
		err = NvBufSurfTransformSetSessionParams (&transform_config_params);
		
		NvBufSurfaceMemSet (dst_surface, 0, 0, 0);
		err = NvBufSurfTransform (surface, dst_surface, &nvbufsurface_params);
		if (err != NvBufSurfTransformError_Success) 
		{
		   g_print ("Error: NvBufSurfTransform failed with error %d while converting buffer\n", err);
		}

		NvBufSurfaceMap (dst_surface, 0, 0, NVBUF_MAP_READ);
		NvBufSurfaceSyncForCpu (dst_surface, 0, 0);
		printf("over NvBufSurfaceSyncForCpu!\r\n");

		need_snap_counter--;
		char currentfolder[64] = {0};
		//nv_dump_dmabuf_to_cpu_memory(dmabuf_fd,0,&templen,(char*)snap_yuv_img.data);
		//nv_dump_dmabuf_to_cpu_memory(dmabuf_fd,1,&templen,(char*)snap_yuv_img.data);    
		memcpy((char*)snap_yuv_img.data,dst_surface->surfaceList[0].mappedAddr.addr[0],4000*3000*3/2);

		printf("over memocpy!\r\n");
		memset(&pub_jpeg_header,0,sizeof(pub_jpeg_header));                    
		get_snap_timestr(currentfolder);
		make_exifparam(&pub_jpeg_header, currentfolder, snap_savepath, index);                       
		is_snap_process = 1;
		if (NvBufSurfaceUnMap (dst_surface, 0, 0))
		{
			printf("Error:NvBufSurfaceUnMap fail!\r\n");
		}
		
		if (NvBufSurfaceDestroy (dst_surface))
		{
			printf("Error:NvBufSurfaceDestroy fail\r\n");
		}
	 }
}
gst_buffer_unmap(gstbuffer, &map);

}

[DaneLLL]

Hi,
Frame data in NV12 has two surfaces. One is Y plane with 4000x3000 bytes. The other is UV plane with 4000x3000x0.5 bytes. Looks like you only map oone surface instead of two.

[blstcvLiu]

Hello,DaneLLL
I had modify the demo to test two sufaces,but there was wrong with it’s width and it’s height,do you have examples of NV12 conversions?

[DaneLLL]

Hi,
Please call

NvBufSurfaceMap (dst_surface, 0, 1, NVBUF_MAP_READ);

and read data from dst_surface->surfaceList[0].mappedAddr.addr[1].

There are patches about creating buffer and call NvBufSurfaceTransform(). Please take a look at
How to include opencv to deepstream-test5-app - #14 by DaneLLL
How to crop the image and save - #2 by DaneLLL
RTSP camera access frame issue - #19 by DaneLLL
Uridecodebin with filesink - #13 by DaneLLL

[blstcvLiu]

Thank you for your help,I solved the problem your way.
Are these processes thread safe？I want to use them in multiple threads.

[DaneLLL]

Hi,
If you will access one buffer in multiple threads, may try to call NvBufSurfTransformAsync() and NvBufSurfTransformSyncObjWait(). So that performance is better. NvBufSurfTransform() is a block operation and may slow other threads.