Auto-StyleMixer: A universal adaptive N-to-One framework for cross-domain data augmentation

Huihuang Zhang, Haigen Hu, Bin Cao, Xiaoqin Zhang

Abstract: Highlights•Proposed Auto-StyleMixer: an N-to-one cross-domain data augmentation framework.•Learns style-mixing configs autonomously, without manual intervention or tuning.•Works with any style extractor, easily plugged into CNNs or Transformers.•Achieves state-of-the-art results on five DG benchmarks.