YOLOv3 doesnt work with deepstream-test2-app

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

I am trying to use a YOLOv3 model as my primary GIE in deepstream-test2-app in deepstream SDK 4.0. I modified the config file as below, but the model does not detect any objects:

[property]
batch-size=1
custom-lib-path=/root/deepstream_sdk_v4.0.2_x86_64/sources/objectDetector_Yolo/nvdsinfer_custom_impl_Yolo/libnvdsinfer_custom_impl_Yolo.so
custom-network-config=/root/deepstream_sdk_v4.0.2_x86_64/sources/objectDetector_Yolo/yolov3.cfg
gie-unique-id=1
gpu-id=0
int8-calib-file=/root/deepstream_sdk_v4.0.2_x86_64/sources/objectDetector_Yolo/yolov3-calibration.table.trt5.1
interval=0
is-classifier=0
labelfile-path=/root/deepstream_sdk_v4.0.2_x86_64/sources/objectDetector_Yolo/labels.txt
maintain-aspect-ratio=1
model-color-format=0
model-file=/root/deepstream_sdk_v4.0.2_x86_64/sources/objectDetector_Yolo/yolov3.weights
net-scale-factor=1
network-mode=1
num-detected-classes=80
parse-bbox-func-name=NvDsInferParseCustomYoloV3
process-mode=1

The pipeline is implemented successfully and the video stream runs smoothly but the number of detected objects remains at 0. The YOLO model itself is set up successfully.

I successfully ran the YOLOv3 model in the deepstream reference app from the config file in objectDetector_Yolo/. I do not know why it does not work with deepstream-test2-app.

I am running the deepstream apps in nvidia-docker containers.

2

Sorry I figured it was because of these 3 lines in the config file:

[class-attrs-all]
threshold=0.2
eps=0.2
group-threshold=1

I commented them out and it started working fine.

I was trying to get it working for almost a week and I finally figured it out, just a few minutes after I posted here.