DANN uses gradient reversal to align the source and the target distribution globally, that are obtained from the deep features of the Convolutional Neural Network (encoder)
cutMix combines two images by retaining features of the source image and the target image in a random ratio.
Combining CutMix with traditional DANN to improve upon the target dataset accuracy for a domain adaptation task. The encoder is jointly trained on the source and cutMix data. This means weights are updated when model is trained on either of the two sets
References -
- Yun, Sangdoo et al. “CutMix: Regularization Strategy to Train Strong Classifiers With Localizable Features.” 2019 IEEE/CVF International Conference on Computer Vision (ICCV) (2019): 6022-6031.
- Y. Ganin, E. Ustinova, H. Ajakan, P. Germain, H. Larochelle, F. Laviolette, M. Marchand, and V. Lempitsky, “Domain-adversarial training of neural networks,” J. Mach. Learn. Res., vol. 17, pp. 59:1–59:35, 2016