Converting from .pth to .onnx failed on 3d-input. Support 3d-conv?

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

Description

Hi,
Below picture is the result of a transformation from pth to onnx who are shown in Neutron:

image
image594×623 39.9 KB

Left is pth’s architecture and right is onnx’s architecture. You noticed that the first layer size of left input is different from the right. In my exprienment, as for the onnx model, the input data flowing from the bottom layer(32x1x3x3x3) which is opppsite from the correct situation (flowing from (320x640x3x3x3)).
My origin network of .pth includes conv3d and the input data is 5-dim. Can the onnx be converted to onnx on conv3d?

Environment

TensorRT Version: 6
GPU Type: 1070
Nvidia Driver Version: 445
CUDA Version: 10.1
CUDNN Version: 7.6.5
Operating System + Version: ubuntu1804
Python Version (if applicable): 3.7.6
TensorFlow Version (if applicable):
PyTorch Version (if applicable): 1.5.1
Baremetal or Container (if container which image + tag):

2

Hi @joevaen,
You may raise this issue on github for better assistance, as this is more of pth to onnx conversion
Issues · onnx/tutorials · GitHub .
Thanks!

3

I cant get my answer straightly in onnx and forum，alas！

4

@AakankshaS i use qat.pth try to convert to onnx but i get the follow error
RuntimeError:
temporary: the only valid use of a module is looking up an attribute but found = prim::SetAttr[name=“num_batches_tracked”](%128, %155)

5

Hi,
Below link might help you with your query, Kindly check below link for all 3d support layers:

Thanks!