Dynamic input shape get wrong result

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

Description

we convert unet from stable-diffusion onnx model with dynamic inputshape
convert command:
./trtexec --onnx=unet.onnx --saveEngine=unet.plan --verbose --fp16 --minShapes=2x4x32x32 --optShapes=2x4x32x32 --maxShapes=2x4x96x96
when inputshape=2x4x40x64 or 2x4x64x64 get the right result but when inputshape=2x4x64x40 get the wrong result

Environment

TensorRT Version: 8.4.3.1
GPU Type: A10
Nvidia Driver Version: 470.129.06
CUDA Version: cuda 11.3
CUDNN Version: 8.05
Operating System + Version: 20.04

2

Hi,

Could you please share with us the minimal issue repro ONNX model and script to try from our end for better debugging.

Thank you.

5

we solve this problem tricky by use multi profile

6

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.