Hi Aditya, this is a good blog. However the `imblearn` library is not meant for image dataset. The fit method have an assertion check of (batch, n_features). Images are 4 dimentional(including batch size).

This reponse is to let you know that maybe the title is misleading. SMOTE is a nice technique but just not practical for images. To use SMOTE with images one need to flatten the image(use permutation invatiant mode) which is not computationally feasible for any practical dataset.

If I am being naive or if I am missing anything do correct me.