My AI Model Is Overfitting Like Crazy – Any Underrated Regularization Hacks?

I'm training a custom image classification model with a relatively small dataset (around 5,000 images per class), and despite using standard dropout and L2 regularization, it's still overfitting terribly on the validation set. I'm hitting a wall here and desperately need some fresh perspectives beyond the usual advice. What unconventional or less-talked-about techniques have genuinely helped you improve generalization in similar scenarios?