Convert ONNX to ENGINE file failure of TensorRT 8.6 when running the command "deepstream-app -c deepstream_app_config.txt" on GPU NVIDIA ORIN NX 64GB

[bravewhh]

Description

When I used the steps provided by Deepstream-YOLO to convert the YOLOv5n model from ONNX to Engine, I received the following error message:

Running Environment

deepstream-app version 6.4.0
DeepStreamSDK 6.4.0
CUDA Driver Version: 12.2
CUDA Runtime Version: 12.2
TensorRT Version: 8.6
cuDNN Version: 8.9
libNVWarp360 Version: 2.0.1d3
TensorRT Version:8.6
NVIDIA GPU: NVIDIA Corporation Device 229e (rev a1)
NVIDIA Driver Version:540.2.0

Operating System:jetson(apply jetpack 6.0)

Running command

python3 export_yoloV5.py -w yolov5s.pt --dynamic
cd Deepstream-yolo
CUDA_VER=12.2 make -C nvdsinfer_custom_impl_Yolo
sudo deepstream-app -c deepstream_app_config.txt

Others

I refer to the issue #2535 and try to generate the onnx model by "python3 export_yoloV5.py -w yolov5s.pt --dynamic --simplify" and "python3 export_yoloV5.py -w yolov5s.pt",but the above question has already exists.......

Please help me to solve the question,thank you so... much !!!

[lix19937]

try to use use fixed shape python3 export_yoloV5.py -w yolov5s.pt --simplify .

[bravewhh]

try to use use fixed shape python3 export_yoloV5.py -w yolov5s.pt --simplify .

I have tried this and the answer is not work.the above question has also existed.

[lix19937]

1. Is it can run x86_64 env ? (with the same nv-cuda-trt version)

2. You can try to use the latest trt version.

[bravewhh]

1. Is it can run x86_64 env ? (with the same nv-cuda-trt version)
2. You can try to use the latest trt version.

sorry，I dont have the x86_64 env with cuda, now the tensorrt running on the jetson env，when i flash the jetpack it include the tensorrt 8.6...

[zerollzeng]

Looks like a dup of #2535

[zerollzeng]

#2535 (comment)