IShuffleLayer applied to shape tensor must have 0 or 1 reshape dimensions: dimensions were [-1,2]

[hanxuanhuo]

Description

I try conver a pytorch model to TensorRT, Pytorch->onnx->TensorRT.
I convert the model to onnx succefully, but when I try to convert onnx to TensorRT using trtexec, I got :
[6] Invalid Node - Pad_14
[shuffleNode.cpp::symbolicExecute::387] Error code 4: Internal Error (Reshape_3: IShuffleLayer applied to shape tensor must have 0 or 1 reshape dimensions: dimensions were [-1,2]

I inspect the model by netron:

It seems the pytorch function F.pad get the incompatib node.
The first step in forward() is :

    h, w = x.shape[-2:]
    extra_h = (math.ceil(w / self.stride[1]) - 1) * self.stride[1] - w + self.kernel_size[1]
    extra_v = (math.ceil(h / self.stride[0]) - 1) * self.stride[0] - h + self.kernel_size[0]
    left = extra_h // 2
    right = extra_h - left
    top = extra_v // 2
    bottom = extra_v - top
    x = F.pad(x, [left, right, top, bottom])
    x = self.conv(x)

Is the problem due to F.pad() function or any other reason?

Environment

TensorRT Version: 8.2.3.0
GPU Type: RTX3060Ti
Nvidia Driver Version: 470.86
CUDA Version: 11.3
CUDNN Version: 8.2.1.32
Operating System + Version:
Python Version (if applicable): 3.7
TensorFlow Version (if applicable):
PyTorch Version (if applicable): 1.10
Baremetal or Container (if container which image + tag):

Relevant Files

Please attach or include links to any models, data, files, or scripts necessary to reproduce your issue. (Github repo, Google Drive, Dropbox, etc.)

Steps To Reproduce

Please include:

• Exact steps/commands to build your repro
• Exact steps/commands to run your repro
• Full traceback of errors encountered

[NVES]

Hi,
Request you to share the ONNX model and the script if not shared already so that we can assist you better.
Alongside you can try few things:

Quick Start Guide :: NVIDIA Deep Learning TensorRT Documentation

1. validating your model with the below snippet

check_model.py

import sys
import onnx
filename = yourONNXmodel
model = onnx.load(filename)
onnx.checker.check_model(model).
2) Try running your model with trtexec command.

TensorRT/samples/trtexec at master · NVIDIA/TensorRT

In case you are still facing issue, request you to share the trtexec “”–verbose"" log for further debugging
Thanks!

[spolisetty]

Hi,

The following may help you.

IShuffleLayer applied to shape tensor must have 0 or 1 reshape dimensions: dimensions were [-1,2])

Hi, I think padding related node is causing error, we don’t support 2D shape tensors yet. We can try workaround constant-fold with polygraphy. After this we are able to successfully generate engine. Please try polygraphy surgeon sanitize --fold-constants grid_sample.onnx -o 2001833/folded.onnx For more details, Thank you.

[hanxuanhuo]

efficientdet.onnx (27.4 MB)
This is my onnx model file.
I can run this by onnxruntime [python] successfully.

[spolisetty]

Hi,

As mentioned in the above post please try the polygraphy tool.
polygraphy surgeon sanitize --fold-constants efficientdet.onnx -o folded.onnx

Thank you.

[hanxuanhuo]

I try this command but get the same problem.
I inspect every step, found the problem is caused by pytorch.onnx.export opset_version.
I got this in opset_version=11:
Constant folding - Only steps=1 can be constant folded for opset >= 10 onnx::Slice op. Constant folding not applied.

This will happen when torch.nn.functional.pad is applied in forward(), polygraphy can not fix this.

[spolisetty]

Thanks for the confirmation. We will check more on this issue.

[spolisetty]

Looks like you marked your previous reply as a solution, do you still need help on this issue?
Thank you.

[hanxuanhuo]

I don’t need this issue anymore. I think I found the problem and the trouble I meet is wait for another one to solve it, like pytorch group or polygraphy group.