Failed to export an ONNX attribute 'onnx::Gather' Bug

[ghost]

🐛 Bug

While trying to convert resnet18 based pytorch model from here. Below error comes up:

RuntimeError: Failed to export an ONNX attribute 'onnx::Gather', since it's not constant, please try to make things (e.g., kernel size) static if possible

To Reproduce

Steps to reproduce the behavior:

1. Download this repo.
   Also download the model weights 79999_iter.pth from the repo page and place it in res/cp.
2. In Google Collab run conversion script:

from model import BiSeNet
import torch.onnx
import torch

net = BiSeNet(19)
net.cuda()
net.load_state_dict(torch.load('/content/drive/My Drive/Collab/fp/res/cp/79999_iter.pth'))
net.eval()

dummy = torch.rand(1,3,512,512).cuda()
torch.onnx.export(net, dummy, "Model.onnx", input_names=["image"], output_names=["output"])

Expected behavior

Should not produce any errors during conversion

Environment

collect_env output:

Collecting environment information...
PyTorch version: 1.4.0
Is debug build: No
CUDA used to build PyTorch: 10.1

OS: Ubuntu 18.04.3 LTS
GCC version: (Ubuntu 7.4.0-1ubuntu1~18.04.1) 7.4.0
CMake version: version 3.12.0

Python version: 3.6
Is CUDA available: Yes
CUDA runtime version: 10.0.130
GPU models and configuration: GPU 0: Tesla T4
Nvidia driver version: 418.67
cuDNN version: /usr/lib/x86_64-linux-gnu/libcudnn.so.7.6.5

Versions of relevant libraries:
[pip3] numpy==1.17.5
[pip3] torch==1.4.0
[pip3] torchsummary==1.5.1
[pip3] torchtext==0.3.1
[pip3] torchvision==0.5.0
[conda] Could not collect

Additional context

I added print (v.node ()) to symbolic_helper.py just before the runtime error is raised to see what's causing the error.

This is the output: %595 : Long() = onnx::Gather[axis=0](%592, %594) # /content/drive/My Drive/Collab/fp/model.py:111:0

And that line in 111 in model.py is: avg = F.avg_pool2d(feat32, feat32.size()[2:])

Based on my further research I found this source stating that:

Both resNet50 and 32 are fine, but to resNet18, ONNX model cannot be exported.

The source suggests the following changes:

From this:

import torch.nn.functional as F
def forward(self, x):
feat = self.base(x)
feat = F.avg_pool2d(feat, feat.size()[2:])

To this:

class Model(nn.Module):
def init():
…
self.avg_pool2d = nn.AvgPool2d(kernel_size=k_s, ceil_mode=False)
…
def forward(self, x):
…
feat = self.avg_pool2d(feat, feat.size()[2:])
…

This change however yield other errors

cc @houseroad @spandantiwari @lara-hdr @BowenBao @neginraoof

[spandantiwari]

@Malemo - I see you have another issue similar to this #34743. Is this a duplicate of that?

[juimoisnono]

l have just solved a similar problem like yours, and i suggest you change "feat", "feat.size()[2:]"to constant, may be it works

[ysnan]

Elaborate a little bit more.
Change instances of

x = F.avg_pool2d(x, x.shape[2:])

to

x_shape = [int(s) for s in x.shape[2:]]
x = F.avg_pool2d(x, x_shape)

make sure the x_shape is the list of int instead of torch.Size().

[garymm]

Duplicate of #34743.