Cannot export rnn together with vgg.

[poor1017]

Describe the bug

I need export a torch model which consists of vgg and rnn into tensorflow. For this purpose I use PyTorch —> ONNX —> Tensorflow approach.

However, I get the following error message at 'model = prepare(onnx_model)':

ValueError: Input size (depth of inputs) must be accessible via shape inference, but saw value None.

I can export vgg and rnn respectively.

To Reproduce

Here's the full code:

`def export(model_path): # Just check VGG2L class TestModel(torch.nn.Module): def load_model(self, model): self.model = model def forward(self, feat, states): return self.model.enc.forward(feat, states)

# Load model
model = torch.load(model_path, map_location=torch.device('cpu'))
model.eval()

test_model = TestModel()
test_model.load_model(model)

# init input
length = 100
input_x = torch.ones([1, length, 83], dtype=torch.float) * 1.3
h0_in = np.zeros((6,1,512),dtype = float)
c0_in = np.zeros((6,1,512),dtype = float)
h0_in = torch.FloatTensor(h0_in)
c0_in = torch.FloatTensor(c0_in)
states = (h0_in, c0_in)

data_input = (input_x, states)

torch.onnx.export(test_model, data_input, 'encoder.onnx',
                  opset_version=10,
                  verbose=True,
                  do_constant_folding=True,
                  input_names=['input', 'h0_in', 'c0_in'],
                  output_names=['output', 'h0_out', 'c0_out'],
                  dynamic_axes={'input':{1:'sequence'},
                                'output': {1:'sequence'},
                               }
                 )

# load onnx model
onnx_model = onnx.load('encoder.onnx')
model = prepare(onnx_model)
# export pb
model.export_graph('encoder.pb')

`

Python, ONNX, ONNX-TF, Tensorflow version

This section can be obtained by running get_version.py from util folder.

• Python version: 3.7
• ONNX version: 1.8.0
• ONNX-TF version: 1.7.0 (tf-1.x)
• Tensorflow version: 1.15

Additional context

The model is like this:

def forward(self, xs_pad, prev_state): xs_pad = self.vgg(xs_pad) xs_pad, states = self.rnn(xs_pad, prev_state) return xs_pad, states

[chinhuang007]

The shape inference error is typically coming from onnx checker. That indicates the onnx model has some issues. Please try to run a simple test with your onnx file to verify.

`import onnx

onnx_path = 'your_model.onnx'

model = onnx.load(onnx_path) onnx.checker.check_model(model) `

[poor1017]

Hi @chinhuang007

    Sorry for getting back to you late. I tried the verification, and it shows that the onnx has no issues.

    Here is the onnx model link:

https://www.dropbox.com/s/5hk2wpb56e2kn0m/enc.onnx?dl=0

    Could you please have a look?