onnx2trt error
cjzcjz666 opened this issue · 0 comments
Description
Environment
TensorRT Version: 8.4.1.5
ONNX-TensorRT Version / Branch: 8.4
GPU Type: Nvidia Orin
Nvidia Driver Version:
CUDA Version: 11.4
CUDNN Version:
Operating System + Version: Ubuntu 20.04
Python Version (if applicable): 3.8
TensorFlow + TF2ONNX Version (if applicable):
PyTorch Version (if applicable): 1.13.0a0+d0d6b1f2.nv22.10
Baremetal or Container (if container which image + tag):
Relevant Files
Steps To Reproduce
When I was deploying the Mobile-SAM model to my end device Nvidia Orin, I encountered a problem while transitioning from the onnx model to the TensorRT model. Error: [TRT] [E] 4: [network. cpp:: validate:: 3032] Error Code 4: Internal Error (orig_im_size: optimization profile is missing values for shape input).
The specific steps to obtain the ONNX model can be referred to https://github.com/ChaoningZhang/MobileSAM ONNX Export
When using the mobile_sam2.onnx file converting to trt, the above error was encountered, with the mentioned in the issue orig_im_size is a fixed static input with a shape of [2]. I don't know why optimization profile is needed.
The python scripts used and my onnx model for onnx2trt are in the relevant files.