DeepStreamSDK API FAQ

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

I am developing a plugin for DeepStream SDK.
DeepStream SDK dsexample_lib I am referring to it.
/opt/nvidia/deepstream/deepstream-4.0/sources/gst-plugins/gst-dsexample
I am using Jetson Nano.
I am using a file in the H.264 encoded yuv420 color format.
The sample plug-in is executed with the following command.
$ deepstream-app -c source8_1080p_dec_infer-resnet_tracker_tiled_display_fp16_nano.txt
The following is added to the config file.

[ds-example]
enable=1
processing-width=1920
processing-height=1080
full-frame=0
unique-id=15
gpu-id=0

Ask about the “sourface” value at line 633 of gstdsexample.cpp

I want to read Y pixel data.
Q1) Should I refer to surface-> surfaceList-> mappedAddr-> addr [0]?
I understand that I use the NvSurfaceMap () API to refer to surface-> surfaceList-> mappedAddr-> addr [0].

Ask about the value of surface-> surfaceList-> layout on line 633 of gstdsexample.cpp.
Values are NVBUF_LAYOUT_PITCH orNVBUF_LAYOUT_BLOCK_LINEAR.
Q2）Please tell me the order of pixel data for NVBUF_LAYOUT_PITCH and NVBUF_LAYOUT_BLOCK_LINEAR. (NV12)

3

Can this page help you?

4

Thank you for your answer.
I was able to access the Y pixel data.
Please tell the order of arrangement of NV12 memory array (Y pixels).
ex) array[ ]={Y1,Y2,Y3…Y1920}

5

you can refer yuv - Image formats NV12 storage in memory - Stack Overflow

6

Thank you for the information.
See the source code answered by Chris Ding.

The order of Ypicel data in the memory below is not arranged line by line.
surface->surfaceList[frame_meta->batch_id].mappedAddr.addr[num_planes]

I was expecting “For a n-pixel NV12 frame: Y × 8 × n (UV) × 2 × n”, but it was different.
Y pixel data is not arranged in order.（array[ ]={Y1,Y2,Y3----Y1920,Y1921-----Y2073600}：for 1080p）
※I understand that I access the data by looking at the padding information.

Please tell me the arrangement order of Y pixel data of mappedAddr.addr [0]. (NV12)

7

It is pitch aligned, you need to note pParams->bytesPerPix[num_planes]*pParams->width[num_planes] <= pParams->pitch[num_planes].