Segmentation Fault during ONNX to TRT Conversion

AI & Data Science Deep Learning (Training & Inference) TensorRT
,
#1

Dear Nvidia Team,

I am Converting my ONNX Model to TRT Model using trtexec, but I am getting an error as Segmentation Fault.
May I know why I am getting that error and how to resolve that? I have attached the Image containing the short Verbose and the error below.

Screenshot from 2021-02-23 03-22-53
Screenshot from 2021-02-23 03-22-531920×1080 487 KB

Command used: ./trtexec --onnx=/home/darshan/slow_fast_1.onnx --shapes=input1:1x3x8x256x455,input2:1x3x32x256x455 --plugins=/home/darshan/RoI_Align.so --plugins=/home/darshan/Einsum_op_dynamic.so --verbose --fp16 --dumpOutput

Environment :

TRT Version: 7.1.3
CUDA: 10.2
OS: Ubuntu 18.04.
GPU: RTX 2060.

Thanks,
Darshan

#2

Looking forward to your reply.

Thanks in Advance,
Darshan

#3

Hi @darshancganji12,

From verbose logs couldn’t get the reason. Segment fault does not tell much about why it happened.
Could you please use onnx runtime and check whether model is valid.

Thank you.

#4

Hi @spolisetty,

Thanks for your reply.
I had checked with ONNX Runtime. The model is valid(I had checked with onnx.checker.check_model(model))

Thanks