How to use SMOTE to rebalance multiclass dataset when the target is one hot encoded with pd.get_dummies?

I’m using a multiclass dataset (cic-ids-2017), which is very imbalanced. I have already encoded the categorical feature (which is the target) using OneHotEncoder.
I tried to use SMOTE oversampling method to balance the data with pipeline:

X = df.drop(['Label'],1)
y = df.Label

steps = [('onehot', OneHotEncoder()), ('smt', SMOTE())]
pipeline = Pipeline(steps=steps)

X, y = pipeline.fit_resample(X, y)

When I used pd.get_dummies instead of OneHotEncoder, in this case I could not use the pipeline (because of get_dummies).

How can I balance the dataset using SMOTE ? (and use get_dummies for One Hot Encoding)