using SIE on full image

Hi,
I have run SSD detector and get the bbox list.
now I want to run SIE (landmark detector) on the frame that is permutation of full frame and the bbox list (that I get from the SSD) to get the landmark of every face.
the problem is that the SIE of landmark is not expect to get the full frame, but struct that I need to build from Full Frame + bbox list.
but the memory aloocation is taken from the .prototext
so the image that is passed to the SIE is not the full frame, but frame that is at the size from the prototxt
so how to handle that problem?

thanks
vadim

Can you try to set “process-mode=1” in nvinfer config

gstnvinfer_property_parser.cpp

switch (val) {
        case 1:
          nvinfer->process_full_frame = TRUE;
          break;
        case 2:
          nvinfer->process_full_frame = FALSE;
          break;

I have seen that , but the problem is that the sizes in the prototxt are not the size of the full frame, and the sizes that are taken to the nvinfer are taken from the prototxt
so I am not processing full frame even if I set that control param to b TRUE

What’s your streammux width/height ?

the stream mux is set to : 1280 x 720
but the input tensor from the prototxt is 64 x 64

You can refer to test2 in default to check your modified code. I think the buffer size sgie received is the full frame size.

gst_pad_push (GST_BASE_TRANSFORM_SRC_PAD (nvinfer), inbuf);

Check your “inbuf” in gst_pad_push() ?