Abstract
We present an unsupervised model without any assumptions to enhance images in nonlinear optical microscopy. It only takes 30 training images and can be generalized to unseen samples. Qualitative and quantitative results show significant improvement.