Learnable Data Augmentation for One-Shot Unsupervised Domain Adaptation
This addresses a challenging domain adaptation scenario for machine learning applications where labeled target data is extremely scarce, representing an incremental improvement over existing methods.
The paper tackles the One-Shot Unsupervised Domain Adaptation problem, where only one unlabeled target sample is available, by introducing a learnable data augmentation method that makes source data perceptually similar to the target, achieving state-of-the-art performance on benchmarks like DomainNet and VisDA.
This paper presents a classification framework based on learnable data augmentation to tackle the One-Shot Unsupervised Domain Adaptation (OS-UDA) problem. OS-UDA is the most challenging setting in Domain Adaptation, as only one single unlabeled target sample is assumed to be available for model adaptation. Driven by such single sample, our method LearnAug-UDA learns how to augment source data, making it perceptually similar to the target. As a result, a classifier trained on such augmented data will generalize well for the target domain. To achieve this, we designed an encoder-decoder architecture that exploits a perceptual loss and style transfer strategies to augment the source data. Our method achieves state-of-the-art performance on two well-known Domain Adaptation benchmarks, DomainNet and VisDA. The project code is available at https://github.com/IIT-PAVIS/LearnAug-UDA