README

 Prerequisites:
- DeepStreamSDK 5.0
- Python 3.6
- Gst-python

 Test Enviromen:
- Jetson NX

 Running Prerequisites:
- Put yolov3.weights to /opt/nvidia/deepstream/deepstream/sources/deepstream_python_apps/apps/deepstream-yolov3-python,weights file can download with darknet official website:https://pjreddie.com/media/files/yolov3.weights
- Compile so library:
-  put diretory nvdsinfer_custom_impl_Yolo in /opt/nvidia/deepstream/deepstream/sources/objectDetector_Yolo,modify nvdsparsebbox_Yolo.cpp 


static const int NUM_CLASSES_YOLO = 80;

then :

export CUDA_VER=10.2

make

copy libnvdsinfer_custom_impl_Yolo.so in /opt/nvidia/deepstream/deepstream/sources/deepstream_python_apps/apps/deepstream-yolov3-python/nvdsinfer_custom_impl_Yolo


modify config_infer_primary_yoloV3.txt


[property]
gpu-id=0
net-scale-factor=0.0039215697906911373
#0=RGB, 1=BGR
model-color-format=0
custom-network-config=yolov3.cfg
model-file=yolov3.weights
model-engine-file=model_b1_gpu0_fp16.engine
labelfile-path=labels.txt
int8-calib-file=yolov3-calibration.table.trt7.0
## 0=FP32, 1=INT8, 2=FP16 mode
network-mode=2
num-detected-classes=80
gie-unique-id=1
network-type=0
is-classifier=0
## 0=Group Rectangles, 1=DBSCAN, 2=NMS, 3= DBSCAN+NMS Hybrid, 4 = None(No clustering)
cluster-mode=2
maintain-aspect-ratio=1
parse-bbox-func-name=NvDsInferParseCustomYoloV3
custom-lib-path=nvdsinfer_custom_impl_Yolo/libnvdsinfer_custom_impl_Yolo.so
engine-create-func-name=NvDsInferYoloCudaEngineGet
#scaling-filter=0
#scaling-compute-hw=0

[class-attrs-all]
nms-iou-threshold=0.3
threshold=0.7

 To run:
  $ python3 deepstream_test_3.py <uri1> [uri2] ... [uriN]
  
e.g.
  $ python3 deepstream_test_3.py file:///home/ubuntu/video1.mp4 file:///home/ubuntu/video2.mp4
  
  $ python3 deepstream_test_3.py rtsp://127.0.0.1/video1 rtsp://127.0.0.1/video2
  
 more info can visit my blogs:https://blog.csdn.net/FL1623863129/article/details/111034113