TGDM: Target Guided Dynamic Mixup for Cross-Domain Few-Shot Learning

Abstract

Given sufficient training data on the source domain, cross-domain few-shot learning (CD-FSL) aims at recognizing new classes with a small number of labeled examples on the target domain. The key to addressing CD-FSL is to narrow the domain gap and transferring knowledge of a network trained on the source domain to the target domain. To help knowledge transfer, this paper introduces an intermediate domain generated by mixing images in the source and the target domain. Specifically, to generate the optimal intermediate domain for different target data, we propose a novel target guided dynamic mixup (TGDM) framework that leverages the target data to guide the generation of mixed images via dynamic mixup. The proposed TGDM framework contains a Mixup-3T network for learning classifiers and a dynamic ratio generation network (DRGN) for learning the optimal mix ratio. To better transfer the knowledge, the proposed Mixup-3T network contains three branches with shared parameters for classifying classes in the source domain, target domain, and intermediate domain. To generate the optimal intermediate domain, the DRGN learns to generate an optimal mix ratio according to the performance on auxiliary target data. Then, the whole TGDM framework is trained via bi-level meta-learning so that TGDM can rectify itself to achieve optimal performance on target data. Extensive experimental results on several benchmark datasets verify the effectiveness of our method.

Deep Learning for Cross-Domain Few-Shot Visual Recognition: A Survey

“…Second, it divides the modulation parameters into the domain-specific and the domain-cooperative sets to explore the intra-domain information and inter-domain correlations, respectively. Furthermore, [139] explores a novel target guided dynamic mixup (TGDM) framework to generate the intermediate domain images to help the FSL task learning on the traget domain. In addition, [76] learns the meta-learners by utilizing multiple domains, and the meta-learners are combined in the parameter space to be the Initialized parameters of a network used in the target domain.…”

Meta-Learning Initialization vs Optimizer: Beyond 20 Ways Few-Shot Learning

“…Contrastive learning approaches are also used, specially in combination with feature selection and with a mixup module that uses a few samples of the target data for image diversity [16], and feature disentanglement to reduce domain bias [20]. A similar idea introduces an intermediate domain created by mixing source and target domain images to bridge the domain gap [3,66]. While prior works, including the ones that evaluate on chest X-ray datasets [16,58], operate in a multi-class setup, we go beyond and consider a multi-label setup with an overlapping train-test label space and domain discrepancies between training and testing.…”