This paper tackles the recent problem of one-shot unsupervised domain adaptation and proposes a new model called Adversarial Style Mining (ASM) to solve it aimed at generating various stylized images for the benefit of downstream tasks (segmentation or classification). After the rebuttal, there is a large consensus that the work has some interesting aspects such as novelty, technical soundness, motivations and valuable results. It was missing a more thourogh discussion about the several steps of the method and comparisons wrt other related approaches (e.g., domain randomization), which seem to be partially addressed by the rebuttal. In the final version, authors are encouraged to address more carefully the reviewers' final remarks and suggestions.