Compile error about facelib:

[wilicyy]

hi,
my box is jetson nx with ds6.0.
i try to make transfer_learning, deepstream_app and faciallandmark together, all is .cpp files.
in my pipline, facenet model is SGIE1 and faciallandmark model is SGIE2.
it builds and makes sucessfully, but when i run my custom app, it shows error to lost the lib file:
./ds-sgs-app: error while loading shared libraries: libnvcv_faciallandmarks.so: cannot open shared object file: No such file or directory,
but i makes that already and pls help to check where is problem?
thank you very much!

this is my make file:
LIBS+= -L$(LIB_INSTALL_DIR) -lnvdsgst_meta -lnvds_meta -lnvdsgst_helper -lnvds_inferutils
-L/usr/local/cuda-$(CUDA_VER)/lib64/ -lcudart -lcuda
-L/opt/nvidia/deepstream/deepstream/lib/cvcore_libs
-lnvcv_faciallandmarks -lnvcv_core -lnvcv_tensorops -lnvcv_trtbackend
-lnvds_msgbroker -lnvds_batch_jpegenc -lnvdsgst_smartrecord -lnvds_utils -lm -lstdc++
-lnvbufsurface -lnvbufsurftransform
-lgstrtspserver-1.0 -lpthread -ldl -Wl,-rpath,$(LIB_INSTALL_DIR)

[wilicyy]

sorry forget that issue pls, i forgot export lib and waste two days time…
and app can runs right now.
but pls give an advice how to add a probe same like sgie_pad_buffer_probe in deepstream_app.cpp?
which place should i put the probe in deepstream_app then it can output tensor of sgie?

thank you very much!

[AastaLLL]

Hi,

The sample creates a pipeline with the create_pipeline(.) function.

The function is implemented in the sample_apps/deepstream-app/deepstream_app.c.
It is the same source of deepstream-app.

Thanks.

[wilicyy]

pls give more advices or informations to add sgie_pad_buffer_probe in deepstream-app pipline, i can not find where can be put the probe…
and another question, faciallandmark-app put few probes on src of next quene, how can i do same place in deepstream-app?

thank you very much!

[wilicyy]

hi,
i am confused on sgie_pad_buffer_probe still where to put in deepstream-app, i tried to instead of gie_processing_done_buf_prob but no face points to draw at all, i think it is the right place already… i added other two probes successuflly in deepstream-app, and only problem for sgie_probe…
pls give one advice soon,

thank you very much!

[wilicyy]

hi,
i add sgie_pad_buffer_probe which copy from facillandmakrs-app to src pad of sgie in deepstream-app,
but happenned segfault.
i think src of sgie should be correct place, pls help to check where is problem? and i have complied deepstream_faciallandmark_meta.cpp together.

  *sink_elem = pipeline->common_elements.secondary_gie_bin.bin

//
NVGSTDS_ELEM_ADD_PROBE (pipeline->common_elements.
primary_bbox_buffer_probe_id,
pipeline->common_elements.secondary_gie_bin.bin, “src”,
sgie_pad_buffer_probe, GST_PAD_PROBE_TYPE_BUFFER,
pipeline->common_elements.appCtx); ///

thank you very much!

[wilicyy]

this is log info and pls help to check where is wrong?

thank you very much

[wilicyy]

[wilicyy]

hi,
i would like to ask again that nvdsnalitics plugin is after sgie or before that?

thank you very much!

[kayccc]

Sorry for the late response, is this still an issue to support? Thanks

[wilicyy]

yes it is still prolems and i am studying insightface and facenet source codes recently.
i would like to ask, nvdsanalysis plugin is before sgie or after sgie in deepstream-app.c?
i am still confused for how to get sgie tensor out at deepstream-app right now.
i have to use face detection as sgie, and recognition by sgie, too.
i used fd-lpd model to detect face.
pls give more advice to get sgie tensor out at deepstream-app.
i tried also NVDIA’s ficaillandmakrs model in deepstream-app pipline, but it is frozed always.

thank you very much!

[AastaLLL]

Hi,

The nvdsanalytics plugin analytics the metadata from the primary detector.
So it should follow by the pgie.

For the tensor issue, have you checked the deepstream-infer-tensor-meta-test sample?
Thanks.

[wilicyy]

so analytics plug in is after pgie and before sgie at deepstream-app, right?
and how can i check all plugins’ order of deepstream-app?
i checked tensor-meta sample, yes.
my problem is how to put probe after sgie in deepstream-app pipline.
i tried here, but failed:

  *sink_elem = pipeline->common_elements.secondary_gie_bin.bin

//
NVGSTDS_ELEM_ADD_PROBE (pipeline->common_elements.
primary_bbox_buffer_probe_id,
pipeline->common_elements.secondary_gie_bin.bin, “src”,
sgie_pad_buffer_probe, GST_PAD_PROBE_TYPE_BUFFER,
pipeline->common_elements.appCtx); ///

[wilicyy]

or give me an example how to insert probe in create pipline(),
thank you very much!

[AastaLLL]

Hi,

For example, in /opt/nvidia/deepstream/deepstream-6.0/sources/apps/sample_apps/deepstream-infer-tensor-meta-test/deepstream_infer_tensor_meta_test.cpp

if(prop.integrated) {
    if (!gst_element_link_many (streammux, pgie, queue, sgie1, queue5,
          sgie2, queue6, sgie3, queue2, tiler, queue3, nvvidconv, queue4, nvosd,
          transform, sink, NULL)) {
    g_printerr ("Elements could not be linked. Exiting.\n");
    return -1;
}
...
tiler_sink_pad = gst_element_get_static_pad (tiler, "sink");
gst_pad_add_probe (tiler_sink_pad, GST_PAD_PROBE_TYPE_BUFFER,
    sgie_pad_buffer_probe, NULL, NULL);

Thanks.

[wilicyy]

i have got sgie tensor out correctly by customize analytics_probe.

thank you.

[wilicyy]

i works on deepstream-app.c and transfer_learning app only,
maybe after tiler_tee i can get same result, too. as there is a good result to use already so i did not test by other ways.
anyway thank you a lots and i can add any probe at anywhere in deepstream-app.c right now.