Deepstream NvBufSurface memory type not changing (dGPU)

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

Hello,

I’m creating a deepstream pipeline on my dGPU, and I’m trying to access the input frames and modify them (without opencv, I’m directly chaning the pixel values), this is part of the pipeline:

nvinfer → nvvideoconvert (nvbuf-memory-type = 3)

To access and edit the frames, I installed a probe at the source pad of the nvvideoconvert plugin:

GstPadProbeReturn srcProbe(GstPad *pad, GstPadProbeInfo *info) {
  GstBuffer *inbuf;
  GstMapInfo in_map_info;

  NvBufSurface *surface = NULL;
  NvDsBatchMeta *batch_meta = NULL;
  NvDsFrameMeta *frame_meta = NULL;
  NvDsMetaList * l_frame = NULL;
  guint i = 0;

  /*CHECK_CUDA_STATUS (cudaSetDevice (0), // gpu_id
      "Unable to set cuda device");*/

  memset (&in_map_info, 0, sizeof (in_map_info));

  // All this following code is preparing the frame buffer to CPU space
  inbuf = GST_PAD_PROBE_INFO_BUFFER(info);
  inbuf = gst_buffer_make_writable(inbuf);

  if (!gst_buffer_map (inbuf, &in_map_info, GST_MAP_WRITE)) {
    g_print ("Error: Failed to map gst buffer\n");
    return GST_PAD_PROBE_OK;
  }

  gst_buffer_unmap(inbuf, &in_map_info);

  surface = (NvBufSurface *) in_map_info.data;

  NvBufSurfaceMap(surface, -1, -1, NVBUF_MAP_WRITE);

  // Modify the frame

  NvBufSurfaceUnMap (surface, 0, 0);


  return GST_PAD_PROBE_OK;
}

Even though the nvvideoconvert plugin is set to memory type 3, when I try to access the surface I get this error:

nvbufsurface: mapping of memory type (2) not supported

Any help would be greatly appreciated, thank you!

• Hardware Platform: GPU
• DeepStream 6.2

3

please refer to this sample for how to access nvBufSurface.

4

I’ve looked at the sample, and mapping the surface in read mode like the sample code actually works:

NvBufSurfaceMap(surface, -1, -1, NVBUF_MAP_READ);

But I need to map it in NVBUF_MAP_WRITE mode, which causes the memory type error I specified earlier.

Should I map the surface in Read, then create a new nvbufsurface infer_buf and map it in read-write mode, modify it, then copy its content into the original with NvBufSurfTransform (inter_buf, surface, &transform_params); ?

5

from the error, NvBufSurfaceMap failed because memory type (2) not supported on dgpu.
please refer NvBufSurfaceMap sample opt\nvidia\deepstream\deepstream-6.2\sources\gst-plugins\gst-dsexample\gstdsexample_optimized.cpp, NVBUF_MEM_CUDA_UNIFIED is supported on dgpu.

6

Found the issue, it looks like I need to change the memory type or resolution of the nvvideoconvert plugin for the memory-type change to work, this needs to be fixed.

7

Sorry for the late reply, Is this still an DeepStream issue to support? Thanks

8

No problem, yes it is an issue with the deepstream plugin Gst-nvvideoconvert whereby if the input format/resolution of the plugin is the same as that of the output the memory-type conversion won’t take place.

Currently, a workaround for this changing either the resolution or the format of the output. You can mark this issue as resolved but this still needs to be fixed in Deepstream.

1 Like
9

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.