This repository provides a custom implementation of parsing function to DeepStream for YOLO models exported using custom Plugins EfficientNMSX_TRT and ROIAling_TRT for Instance Segmentation Models and EfficientNMS_TRT for Detection Models.
NvDsInferYoloMask
- For Segmentation Models
NvDsInferYoloNMS
- For Detection Models
To clone the repository and build\install the libnvds_infer_yolo_nms.so
library, follow these steps:
git clone https://github.com/levipereira/nvdsinfer_yolo_nms.git
cp -R nvdsinfer_yolo_nms/ /opt/nvidia/deepstream/deepstream/sources/libs/
cd /opt/nvidia/deepstream/deepstream/sources/libs/
CUDA_VER=12.2 make -C nvdsinfer_yolo_nms install
Install Location:
/opt/nvidia/deepstream/deepstream/lib/libnvds_infer_yolo.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: "NvDsInferYoloNMS"
}
}
custom_lib {
path : "/opt/nvidia/deepstream/deepstream/lib/libnvds_infer_yolo.so"
}