Deepstream app on python use yolo3 as primary infer

Hello everybody, want to use yolo3 as primary infer in DS Python APP.
I compile an objectDetoctorYolo from source folder DS. On config add line:

output-blob-names=yolo_83;yolo_95;yolo_107

App successfully started, but I have an empty l_obj var and not detected objects.

what am I doing wrong? Thanks, platform: Jetson Xavier AGX

You should have your own yolo cfg file like yolov3.cfg
“output-blob-names” is useless except you use engine file directly.

To use the objectDetectorYolo sample with your python app, try this config in place of your dstestx_pgie_config.txt. Make sure to remove the [class-attr-all] section with the clustering configs. This is basically the same as config_infer_primary_yoloV3.txt with paths adjusted assuming your python app is in the standard location.

[property]
gpu-id=0
batch-size=1
interval=0
gie-unique-id=1
nvbuf-memory-type=0
config-file=config_infer_primary_yoloV3.txt
net-scale-factor=1
#0=RGB, 1=BGR
model-color-format=0
custom-network-config=../../../objectDetector_Yolo/yolov3.cfg
model-file=../../../objectDetector_Yolo/yolov3.weights
model-engine-file=model_b1_int8.engine
labelfile-path=../../../objectDetector_Yolo/labels.txt
int8-calib-file=../../../objectDetector_Yolo/yolov3-calibration.table.trt5.1
## 0=FP32, 1=INT8, 2=FP16 mode
network-mode=1
num-detected-classes=80
gie-unique-id=1
is-classifier=0
maintain-aspect-ratio=1
parse-bbox-func-name=NvDsInferParseCustomYoloV3
custom-lib-path=../../../objectDetector_Yolo/nvdsinfer_custom_impl_Yolo/libnvdsinfer_custom_impl_Yolo.so
1 Like