Primary-gie and secondary-gie problem

[MingGatsby]

• Hardware Platform (Jetson / GPU): Xavier NX
• DeepStream Version: 5.0.1
• JetPack Version (valid for Jetson only): 4.5
• TensorRT Version: 7.1.3
• Issue Type( questions, new requirements, bugs): questions

Integrated deepstream-pose-estimation into deepstream-app with smalll display issue

Hi The main steps are as follows: In deepstream-app main(), I add a probe at the src pad of secondary gie since I use primary gie with a yolo detector: GstPad *src_pad = NULL; GstElement *pose = appCtx[i]->pipeline.common_elements.secondary_gie_bin.sub_bins[0].secondary_gie; src_pad = gst_element_get_static_pad (pose, "src"); if (!src_pad) g_print ("Unable to get primary_gie src pad\n"); else { gst_pad_add_probe(src_pad, GST_PAD_PROBE_TYPE_BUFFER, pgie_…

Hi everybody,

I need to combine pose estimation and yolo on the deepstream now. I already ref the article, and I can use yolo or pose estimation Individually.

Nevertheless, I still don’t understand how to combine them.
Whether to use primary_gie and secondary-gie0 for solving the problem?
Because I add pose estimation to [primary-gie] config, and yolo to [secondary-gie0] config, but secondary-gie0 don’t run successfully.

Or can I use two primary_gie for solving the problem?
Can I ref which problem or article? I can’t understand the using methods of [primary-gie] between [secondary-gie0].

The following is my code and the config.

code.zip (9.7 KB)

[AndySimcoe]

Hi,

In terms of “combine”, you mean cascading two GIEs, one for yolo and the other for pose-estimation and not retrieving yolo’s bbox info for the pose-estimation? Because this is the case I did.

In this scenario, you can use either GIEs for yolo or pose-estimation and get meta data by adding two probes on relevant src-pad.

Maybe you forgot to enable “output-tensor-meta” for the pose-estimation GIE?

[MingGatsby]

Hi,

In terms of “combine”, you mean cascading two GIEs, one for yolo and the other for pose-estimation.
====>Yes, this my meaning.

get meta data by adding two probes on relevant src-pad.
Maybe you forgot to enable “output-tensor-meta” for the pose-estimation GIE?
=====>I understand I can add them to test1~5.app example.
I already ref your article code, but it’s deepstream.app.I don’t understand how to add them to where.

You can set this property in “deepstream_primerary(or secondary)_gie_bin.c” file or you can set it in deepstream-app config file.
===> Sorry, I can’t understand your meaning, how to set it in deepstream-app config file?
I already ref /opt/nvidia/deepstream/deepstream-5.0/samples/configs/deepstream-app/source4_1080p_dec_infer-resnet_tracker_sgie_tiled_display_int8.txt

But I still can’t use yolo and pose estimation together. The following is my config, please check it , thank you.

code.zip (171.3 KB)

[MingGatsby]

Maybe you forgot to enable “output-tensor-meta” for the pose-estimation GIE?
====>
Actually, I don’t understand how to enable “output-tensor-meta” for the pose-estimation GIE on deepstream-app config file. If you agree, would you mind to share your config and code about “output-tensor-meta”.

The following image is correct? But I still only primary_gie output, can’t see secondary_gie. I sure I aready update to .o file.

OH~ I success, I change process-mode to “1”, thank you so much!!

[MingGatsby]

I have a small question, I can use deepstream_secondary_gie_bin.c to solve problem.
However , I wanna use config to set, it can’t be success.

output-tensor-meta=1
process-mode=1

After I test that, “output-tensor-meta” on config is running, but “process-mode” on config isn’t running. I don’t know why. @@

Therefore, I still must fix deepstream_secondary_gie_bin.c to “process-mode” 1.

I fix deepstream_secondary_gie_bin.c, it will be success to use “process-mode” on config.