Target-Aware Generative Augmentations for Single-Shot Adaptation
This addresses the brittle generalization of deep neural networks in domain adaptation, particularly for single-shot scenarios, offering a method to handle large distribution shifts, though it is incremental as it builds on existing test-time adaptation techniques.
The paper tackles the problem of adapting models from a source to a target domain with limited data, proposing SiSTA, which fine-tunes a generative model using a single target sample and uses novel sampling strategies to improve generalization, achieving significant gains in face attribute detection and multi-class object recognition, with performance competitive to models trained on larger datasets.
In this paper, we address the problem of adapting models from a source domain to a target domain, a task that has become increasingly important due to the brittle generalization of deep neural networks. While several test-time adaptation techniques have emerged, they typically rely on synthetic toolbox data augmentations in cases of limited target data availability. We consider the challenging setting of single-shot adaptation and explore the design of augmentation strategies. We argue that augmentations utilized by existing methods are insufficient to handle large distribution shifts, and hence propose a new approach SiSTA, which first fine-tunes a generative model from the source domain using a single-shot target, and then employs novel sampling strategies for curating synthetic target data. Using experiments on a variety of benchmarks, distribution shifts and image corruptions, we find that SiSTA produces significantly improved generalization over existing baselines in face attribute detection and multi-class object recognition. Furthermore, SiSTA performs competitively to models obtained by training on larger target datasets. Our codes can be accessed at https://github.com/Rakshith-2905/SiSTA.