DeepStream Yolo

NVIDIA DeepStream SDK 6.0 configuration for YOLO models

marcoslucianops
405
118
C++

DeepStream-Yolo

NVIDIA DeepStream SDK 7.1 / 7.0 / 6.4 / 6.3 / 6.2 / 6.1.1 / 6.1 / 6.0.1 / 6.0 / 5.1 configuration for YOLO models

For now, I am limited for some updates. Thank you for understanding.

YOLO-Pose: https://github.com/marcoslucianops/DeepStream-Yolo-Pose

YOLO-Seg: https://github.com/marcoslucianops/DeepStream-Yolo-Seg

YOLO-Face: https://github.com/marcoslucianops/DeepStream-Yolo-Face

Important: please export the ONNX model with the new export file, generate the TensorRT engine again with the updated files, and use the new config_infer_primary file according to your model

Improvements on this repository

  • Support for INT8 calibration
  • Support for non square models
  • Models benchmarks
  • Support for Darknet models (YOLOv4, etc) using cfg and weights conversion with GPU post-processing
  • Support for RT-DETR, CO-DETR (MMDetection), YOLO-NAS, PPYOLOE+, PPYOLOE, DAMO-YOLO, Gold-YOLO, RTMDet (MMYOLO), YOLOX, YOLOR, YOLOv9, YOLOv8, YOLOv7, YOLOv6 and YOLOv5 using ONNX conversion with GPU post-processing
  • GPU bbox parser
  • Custom ONNX model parser
  • Dynamic batch-size
  • INT8 calibration (PTQ) for Darknet and ONNX exported models

Getting started

Requirements

DeepStream 7.1 on x86 platform

DeepStream 7.0 on x86 platform

DeepStream 6.4 on x86 platform

DeepStream 6.3 on x86 platform

DeepStream 6.2 on x86 platform

DeepStream 6.1.1 on x86 platform

DeepStream 6.1 on x86 platform

DeepStream 6.0.1 / 6.0 on x86 platform

DeepStream 5.1 on x86 platform

DeepStream 7.1 on Jetson platform

DeepStream 7.0 on Jetson platform

DeepStream 6.4 on Jetson platform

DeepStream 6.3 on Jetson platform

DeepStream 6.2 on Jetson platform

DeepStream 6.1.1 on Jetson platform

DeepStream 6.1 on Jetson platform

DeepStream 6.0.1 / 6.0 on Jetson platform

DeepStream 5.1 on Jetson platform

Supported models

Basic usage

1. Download the repo

git clone https://github.com/marcoslucianops/DeepStream-Yolo.git
cd DeepStream-Yolo

2. Download the cfg and weights files from Darknet repo to the DeepStream-Yolo folder

3. Compile the lib

3.1. Set the CUDA_VER according to your DeepStream version

export CUDA_VER=XY.Z

  • x86 platform

    DeepStream 7.1 = 12.6
DeepStream 7.0 / 6.4 = 12.2
DeepStream 6.3 = 12.1
DeepStream 6.2 = 11.8
DeepStream 6.1.1 = 11.7
DeepStream 6.1 = 11.6
DeepStream 6.0.1 / 6.0 = 11.4
DeepStream 5.1 = 11.1

  • Jetson platform

    DeepStream 7.1 = 12.6
DeepStream 7.0 / 6.4 = 12.2
DeepStream 6.3 / 6.2 / 6.1.1 / 6.1 = 11.4
DeepStream 6.0.1 / 6.0 / 5.1 = 10.2

3.2. Make the lib

make -C nvdsinfer_custom_impl_Yolo clean && make -C nvdsinfer_custom_impl_Yolo

4. Edit the config_infer_primary.txt file according to your model (example for YOLOv4)

[property]
...
custom-network-config=yolov4.cfg
model-file=yolov4.weights
...

NOTE: For Darknet models, by default, the dynamic batch-size is set. To use static batch-size, uncomment the line

...
force-implicit-batch-dim=1
...

5. Run

deepstream-app -c deepstream_app_config.txt

NOTE: The TensorRT engine file may take a very long time to generate (sometimes more than 10 minutes).

NOTE: If you want to use YOLOv2 or YOLOv2-Tiny models, change the deepstream_app_config.txt file before run it

...
[primary-gie]
...
config-file=config_infer_primary_yoloV2.txt
...

Docker usage

  • x86 platform

    nvcr.io/nvidia/deepstream:7.1-gc-triton-devel
nvcr.io/nvidia/deepstream:7.1-triton-multiarch

  • Jetson platform

    nvcr.io/nvidia/deepstream:7.1-triton-multiarch

NOTE: To compile the nvdsinfer_custom_impl_Yolo, you need to install the g++ inside the container

apt-get install build-essential

NOTE: With DeepStream 7.1, the docker containers do not package libraries necessary for certain multimedia operations like audio data parsing, CPU decode, and CPU encode. This change could affect processing certain video streams/files like mp4 that include audio track. Please run the below script inside the docker images to install additional packages that might be necessary to use all of the DeepStreamSDK features:

/opt/nvidia/deepstream/deepstream/user_additional_install.sh

NMS Configuration

To change the nms-iou-threshold, pre-cluster-threshold and topk values, modify the config_infer file

[class-attrs-all]
nms-iou-threshold=0.45
pre-cluster-threshold=0.25
topk=300

NOTE: Make sure to set cluster-mode=2 in the config_infer file.

Notes

  1. Sometimes while running gstreamer pipeline or sample apps, user can encounter error: GLib (gthread-posix.c): Unexpected error from C library during 'pthread_setspecific': Invalid argument. Aborting.. The issue is caused because of a bug in glib 2.0-2.72 version which comes with Ubuntu 22.04 by default. The issue is addressed in glib 2.76 and its installation is required to fix the issue (https://github.com/GNOME/glib/tree/2.76.6).

    • Migrate glib to newer version

      pip3 install meson
pip3 install ninja

      NOTE: It is recommended to use Python virtualenv.

      git clone https://github.com/GNOME/glib.git
cd glib
git checkout 2.76.6
meson build --prefix=/usr
ninja -C build/
cd build/
ninja install

    • Check and confirm the newly installed glib version:

      pkg-config --modversion glib-2.0

  2. Sometimes with RTSP streams the application gets stuck on reaching EOS. This is because of an issue in rtpjitterbuffer component. To fix this issue, a script has been provided with required details to update gstrtpmanager library.

    /opt/nvidia/deepstream/deepstream/update_rtpmanager.sh

Extract metadata

You can get metadata from DeepStream using Python and C/C++. For C/C++, you can edit the deepstream-app or deepstream-test codes. For Python, your can install and edit deepstream_python_apps.

Basically, you need manipulate the NvDsObjectMeta (Python / C/C++) and NvDsFrameMeta (Python / C/C++) to get the label, position, etc. of bboxes.

My projects: https://www.youtube.com/MarcosLucianoTV