I exported a onnx described resnet18 model from pytorch0.4.1 to nnvm, but I got a bit of mistake. It shows that the shape does not match. But I got a onnx model from resnet18 and it works well on nnvm. I used the netron graphics software to examine in detail the differences between the two resnet18 neural networks. Both seem to differ only in the final network layer. As follows:
The above is a successful run of resnet18.onnx when loading into nnvm, the following is resnet18.onnx exported from pytorch0.4.1 and showing that shape does not match after loading into nnvm Well, is it that nnvm does not support these network layers or other issues, I am very confused now.
