Deploy the Yolov3 model trained on TLT, but the target cannot be detected

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

I trained the Yolov3 model on TLT, the ｍAP is 88.9%, and verified that the deployed model can accurately detect the target.
But using the Yolov3 example in DeepStream to deploy the trained etlt model, it runs without problems, but the target cannot be detected. Why is this happening? how should I solve this?

image
image1723×516 22 KB

Thank you very much

Environment

train
TensorRT Version : 7.2.1.6
GPU Type : 3080
Nvidia Driver Version : 460.39
CUDA Version : 10.2
CUDNN Version : 8.1.0
Operating System + Version : Ubuntu18.04
Python Version (if applicable) : 3.6.9
TLT ：3.0
deploy:
TensorRT Version : 7.2.1.6
GPU Type : V100
Nvidia Driver Version : 450.80.02
CUDA Version : 10.2
CUDNN Version : 8.1.1
Operating System + Version : Ubuntu18.04
Python Version (if applicable) : 3.6.9
DeepStream ：5.0.0

3

Did you change any config in the training?
If you did, you need to make the correspding change to the DS inference config file, i.e. deepstream_app_config_yolo_tlt.txt.