NvDsInferYoloEfficientNMS for Gst-nvinferserver

This repository provides a custom implementation of parsing function to the Gst-nvinferserver plugin when use YOLOv7/YOLOv9 model served by Triton Server using the Efficient NMS plugin exported by ONNX.

By using the parsing function provided by NvDsInferYoloEfficientNMS, handling the number of classes dynamically becomes easier. This eliminates the need to hardcode the number of classes, allowing the same plugin to be used for different YOLOv9 models with varying numbers of classes.

• Use parse function NvDsInferYoloEfficientNMS for YOLO Exported ONNX model with End2End

Deployment Guide for NvDsInferYoloEfficientNMS

Cloning Repository and Installation

To clone the repository and build\install the libnvds_infer_yolo_efficient_nms.so library, follow these steps:

# Clone the repository
git clone https://github.com/levipereira/nvdsinfer_yolo_efficient_nms.git

# Copy the repository to the desired location
cp -R nvdsinfer_yolo_efficient_nms/ /opt/nvidia/deepstream/deepstream/sources/libs/

# Set the CUDA_VER environment variable (check your deepstream cuda version.  The DS 6.4 use cuda 12.2)
export CUDA_VER=12.2

# Navigate to the directory containing the nvdsinfer_yolov9_efficient_nms library
cd /opt/nvidia/deepstream/deepstream/sources/libs/nvdsinfer_yolo_efficient_nms

# Build the project using the provided MakeFile
make -f MakeFile all
make -f MakeFile install

Install Location:

/opt/nvidia/deepstream/deepstream/lib/libnvds_infer_yolo_efficient_nms.so

Usage on Deepstream

Snippet Gst-nvinferserver Configuration File for YOLO Models

  postprocess {
    labelfile_path: "labels.txt"
    detection {
      num_detected_classes: 80
      custom_parse_bbox_func: "NvDsInferYoloEfficientNMS"
    }
  }
  custom_lib {
    path : "/opt/nvidia/deepstream/deepstream/lib/libnvds_infer_yolo_efficient_nms.so"
  }