Converting to Onnx gives weird output

[mhrrs]

I used the following command to convert my detectron2 model to onnx (I replaced the path names, but they are correct when I run the command):
python "...\export_model.py" --sample-image "...\00D61B56-086-nc-T_i.pgm" --config-file "...\config.yaml" --export-method tracing --format onnx --output "...\export_model_test_NC_v1" MODEL.WEIGHTS "...\model_final.pth"

This creates the onnx model No problems yet.

I then run the following python code:
`

Load your ONNX model globally on server start

model_path = r'...\model.onnx'
ort_session = ort.InferenceSession(model_path)

Create a dummy input tensor

dummy_input = torch.randn(3, 1333, 332) <---- THIS IS MY ISSUE

Load the ONNX model

ort_session = ort.InferenceSession(model_path)

Run inference

outputs = ort_session.run(None, {"x.1": dummy_input.numpy()})

Print the outputs

print(outputs)
`

For some reason the only way I can get this python code to work is to use those dimensions. I didn't set those anywhere. What is going on? It then results in the following output:
[array([], shape=(0, 4), dtype=float32), array([], dtype=int64), array([], shape=(0, 1, 28, 28), dtype=float32), array([], dtype=float32), array([1333, 332], dtype=int64)]