How to split and crop a box in a landmark_pad_probe nad infer with another element?

[erence]

Please provide the following information when requesting support.

• Hardware (RTX2070)
• Network Type (Yolo_v4_tiny)
• TLT Version (Tao 3.21.11)

Hi. I am using fpenet–>facelandmark model example of deepstream 6.2. Through some landmarks
in GstPadProbeReturn function I define a rectangle like:
disp_meta->rect_params[disp_meta->num_rects].left ==xx top width and height.
and
disp_meta->rect_params[disp_meta->num_rects+1].left etc

I want now to split first into two pipelines that each infer on this squares as seperate roi. How can I split?

Should I then crop? How can I define a src according to this rect_params? Can I retrieve rect_params from GstPadProbeReturn for crop element ?
which element is best with deepstream to crop? Can I have two seperate sinks as display in the end?

[yingliu]

Hi,
This looks more related to Deepstream. We are moving this post to the Deepstream forum to get better help.
Thank you.

[yuweiw]

You can try to use a tee plugin as the split element in your picture. Then use the nvvideoconvert plugin as the crop element.

[erence]

Could you please

@yuweiw

nvvideoconvert

Thank you indeed. tee seprates the pipeline easily. Can you please also indicate how can I crop dynamically using the data in a GstPadProbeReturn (disp_meta->rect_params[disp_meta->num_rects]) . Should I create a new appsrc ? Can I just use crop with rect_params with nvvideoconvert?

[yuweiw]

You can try the mothod below:
https://forums.developer.nvidia.com/t/dynamically-adjusting-crop-to-input-resolution/246828/4

[erence]

Thanks. I receive freeze error when using the method you mentioned. "g_object_set(G_OBJECT(nvvidconv_crop), “src-crop”, scr_crop.c_str(), NULL); How can I free the memoryfrom scr_crop-c_str in pad probe function? Should I not crop with a pad?
g_free does not work.

[yuweiw]

What do you mean I receive freeze error when using the method you mentioned?
About how to free the string, you can refer to the link below:
https://github.com/NVIDIA-AI-IOT/deepstream_reference_apps/blob/master/deepstream-bodypose-3d/sources/deepstream_pose_estimation_app.cpp#L1861

[erence]

The pipeline does not work. It freezes, or gives core dump error.

I suppose I can add data to pad with
gst_pad_probe_function :

g_object_set (G_OBJECT (nvvideoconvert), “src-crop”, x, y, width, height, NULL);

but I cannot retrieve data from Pad to nvvideocenvert with:

nvvideoconvert = makeElementAndLink(“nvvideoconvert”, NULL, bin, element);
GstPad *pad = gst_element_get_static_pad (nvvideoconvert, “src”);
gst_pad_add_probe (pad, GST_PAD_PROBE_TYPE_BUFFER, gst_pad_probe_function , nvvideoconvert, NULL);

i can crop though with

g_object_set(G_OBJECT(nvvideoconvert), “src-crop” , “300:100:100:200”, NULL);

as an example, but cannot change 300:100:100:200 to values, that are written dynamically in gst_pad_probe function

[yuweiw]

There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks

@erence

as an example, but cannot change 300:100:100:200 to values, that are written dynamically in gst_pad_probe function

You can refer to the link I attached before:

  gchar *string1 = NULL;
  asprintf (&string1, "%d:%d:%d:%d", _pad_dim, _pad_dim, _image_width, _image_height);