This PR fixes the bug related to skip_connet in the BottleNeckBlock block.
Instead of using the flattened tensor for getting the original dimension and adding with the output tensor, the original tensor was used, which will cause a shape mismatch error for an original input tensor with more than 2 dimensions.
This PR fixes the bug related to skip_connet in the BottleNeckBlock block.
Instead of using the flattened tensor for getting the original dimension and adding with the output tensor, the original tensor was used, which will cause a shape mismatch error for an original input tensor with more than 2 dimensions.