This website requires JavaScript.

A Survey of Deep Visual Cross-Domain Few-Shot Learning

Wenjian WangLijuan DuanYuxi WangJunsong FanZhi GongZhaoxiang Zhang
Mar 2023
Few-Shot transfer learning has become a major focus of research as it allowsrecognition of new classes with limited labeled data. While it is assumed thattrain and test data have the same data distribution, this is often not the casein real-world applications. This leads to decreased model transfer effects whenthe new class distribution differs significantly from the learned classes.Research into Cross-Domain Few-Shot (CDFS) has emerged to address this issue,forming a more challenging and realistic setting. In this survey, we provide adetailed taxonomy of CDFS from the problem setting and corresponding solutionsview. We summarise the existing CDFS network architectures and discuss thesolution ideas for each direction the taxonomy indicates. Furthermore, weintroduce various CDFS downstream applications and outline classification,detection, and segmentation benchmarks and corresponding standards forevaluation. We also discuss the challenges of CDFS research and explorepotential directions for future investigation. Through this review, we aim toprovide comprehensive guidance on CDFS research, enabling researchers to gaininsight into the state-of-the-art while allowing them to build upon existingsolutions to develop their own CDFS models.