nvidia-ai-iot / deepstream_parallel_inference_app Goto Github PK

Hi,
I have split a single cam source into 4 different ROIs and feed it to inference layer. This solve issue of detecting small objects but How to use gst-nvdsmetamux to fuse adjacent objects from different ROI into one object to track

After seeing the GTC session dedicated to this parallel infererence, first I have to give you my congratulations, and second I would like to know how is it going to work with dynamic pipelines? in all the samples we are using static files, but in my requirements I would need to work with dynamic configurations, do you have some plan on this?

Regards.

I have a custom trained yolov5 model, how can I replace the tao yolov4 with that?

I just followed the steps of commands show in Readme：

1. apt install git-lfs
   git lfs install --skip-repo
   git clone https://github.com/NVIDIA-AI-IOT/deepstream_parallel_inference_app.git
2. apt-get install -y libjson-glib-dev libgstrtspserver-1.0-dev
3. cd tritonserver/
   ./build_engine.sh
   4.cd tritonclient/sample/
   source build.sh
   ./apps/deepstream-parallel-infer/deepstream-parallel-infer -c configs/apps/bodypose_yolo_lpr/source4_1080p_dec_parallel_infer.yml

I can run all the commands sucessfully except the last line, the errors are shown below:

Ideally I want three models ran in parallel but their predictions all visible on a 1x1 tiled display.

Not too fussed about what the models actually are at the moment

Thank You

I try to set different source_id with different model.
I found similar settings in deepstream_app_config.yml and config_metamux.txt。 Here is the snapshoot as below:

Hope someone can help to clarify it, thanks in advance!

Hello, Sir/Madam
I am running the deepsteam-parallel_infer_app in my job now. I want to save the image on Jetson platform.
When I set up multiple steams, I found nvds_obj_enc_process has problem. It saved the wrong source image:
When I execute the Nvds_obj_enc_process function, the frame image saved is the wrong source frame image. I want to know what operations the Nvds_obj_enc_process function performs. Is it the offset error of the current frame that causes the acquired frame image to be incorrect or what is the reason?

How you have obtained libnvds_osd.so file from scrtch

When pulling RTSP video streams, the source is added above 16, there is an issue of not being able to link to the tee. The following is an error message

What is the reason for this？