how to pMapperedBuffer to opencv Mat?

[kimsjpk1]

i tried to apply super resolution in deepstream code

and i watched your github code

example/cpp/4file_custom_pph_using_opencv.cpp

// Transforming only one frame in the batch, so create a copy of the single
// surface ... becoming our new source surface. This creates a new mono
// (non-batched) surface copied from the "batched frames" using the batch id
// as the index
DSL::DslMonoSurface monoSurface(pMappedBuffer->pSurface,
pFrameMeta->batch_id);

        // Coordinates and dimensions for our destination surface.
        int left(0), top(0);

        // capturing full frame
        int width = pMappedBuffer->GetWidth(pFrameMeta->batch_id);
        int height = pMappedBuffer->GetHeight(pFrameMeta->batch_id);

if pMappedBuffer with batch_id convert to opencv Mat, i can apply super resolution on Frame

please tell me how to convert it

thanks in advance

[kimsjpk1]

i search information and found it

https://forums.developer.nvidia.com/t/deepstream-sample-code-snippet/142683

but i want to change buffer before applied objected detection

in deepstream-service-library/src/DslSurfaceTransform.h, i want to change code

DslMappedBuffer(GstBuffer* pBuffer)
: GstMapInfo GST_MAP_INFO_INIT
, m_pBuffer(NULL)
{
LOG_FUNC();

        if (!gst_buffer_map(pBuffer, this, GST_MAP_READ))
        {
            LOG_ERROR("Failed to map gst buffer");
            throw;
        }
        // set only when successful
        m_pBuffer = pBuffer;

        pSurface = (NvBufSurface*)data;
    }

pSurface->surfaceList[index] to cv::Mat and apply super resolution to pSurface -> surfaceList[index]

[rjhowell44]

@kimsjpk1 I'll take a look at this tomorrow (Monday)

[rjhowell44]

@rjhowell44 my apologies for the slow response. After some review, I think I understand what you're looking for. First, DslSurfaceTransform.h was developed to simplify the process of copying a frame to a new buffer surface so it can be mapped and manipulated.

If I understand correctly, you want to update the actual pipeline buffer prior to your inference component.

Looking at the code snippets you provided, this should be possible if done in a custom pph added to the sink pad of the primary inference component. You won't need DslSurfaceTransform.h.

The opencv super resolution looks really interesting, and I can see the value in having an example of this. I'm currently working on #1225 and should be done in a couple of days. If you can give me a few days, I should be able to work on this next.