FS-BAN: Born-Again Networks for
Domain Generalization Few-shot Classification
Yunqing Zhao
Ngai-Man Cheung
Singapore University of Technology and Design (SUTD)
[Paper]
[GitHub]

Overview and our contributions.
- 1: We consider the problem of tackling few-shot classification on domains unseen in training, i.e., domain generalization few-shot classification (DG-FSC).
- 2: Our work makes two main contributions:

  • We discover and propose that the Born-Again Networks, which proposed in conventional supervised learning tasks, are still useful in episodic training for few-shot classification.
  • We propose few-shot born-again networks (FS-BAN), a novel BAN approach that addresses unique issues in DG-FSC. The proposed FS-BAN achieves consistent state-of-the-art performance over different baseline FSC models and six public datasets.
- 3: Schematic diagram of our proposed Few-shot Born-Again Networks (FS-BAN), see Figure in Main Results part: 1: Mutual Regularizaion. 2: Knowledge transfer with mismatched teacher. 3. Meta-control the temperature. The proposed multi-task learning objectives tackles unique challenges in DG-FSC effectively and are verified in carefully designed ablation studies.

Abstract

Conventional Few-shot classification (FSC) aims to recognize samples from novel classes given limited labeled data. Recently, domain generalization FSC (DG-FSC) has been proposed with the goal to recognize novel class samples from unseen domains. DG-FSC poses considerable challenges to many models due to the domain shift between base classes (used in training) and novel classes (encountered in evaluation). In this work, we make two novel contributions to tackle DG-FSC. Our first contribution is to propose Born-Again Network (BAN) episodic training and comprehensively investigate its effectiveness for DG-FSC. As a specific form of knowledge distillation, BAN has been shown to achieve improved generalization in conventional supervised classification with a closed-set setup. This improved generalization motivates us to study BAN for DG-FSC, and we show that BAN is promising to address the domain shift encountered in DG-FSC. Building on the encouraging finding, our second (major) contribution is to propose few-shot BAN, FS-BAN, a novel BAN approach for DG-FSC. Our proposed FS-BAN includes novel multi-task learning objectives: Mutual Regularization, Mismatched Teacher and Meta-Control Temperature, each of these is specifically designed to overcome central and unique challenges in DG-FSC, namely overfitting and domain discrepancy. We analyze different design choices of these techniques. We conduct comprehensive quantitative and qualitative analysis and evaluation using six datasets and three baseline models. The results suggest that our proposed FS-BAN consistently improves the generalization performance of baseline models and achieves state-of-the-art accuracy for DG-FSC.


Investigation: Born-Again Network (BAN) for
Cross-domain FSC in Generations

Born-Again Nets in episodic meta-learning (above) and results of our investigation(below):

Main results: Experiment With Proposed FS-BAN

Overview of proposed method (above) and main results (below), note that there is only one generation, details in our paper.

(Top) Meta-test accuracy (%) of DG-FSC with our proposed FS-BAN. We follow experiment setup as in Tseng et al. We let All={miniImageNet, CUB, Cars, Places, Plantae} is the union of all domains for training and testing. In training phase, we sample tasks from multiple seen domains, e.g., All\ {CUB}. In testing phase, we evaluate the model on tasks sampled from the leave-one-out selected unseen domain, e.g., CUB. miniImageNet is always the source domain. (Bottom) The effectiveness of each component in multi-task learning objectives.


Paper Additional Information

Yunqing Zhao and Ngai-Man Cheung.
FS-BAN: Born-Again Networks for Domain Generalization Few-shot Classification
In IEEE Trans. on Image Processing (T-IP), 2023.
(hosted on arXiv)

If you find our work useful in your research, please consider citing our paper: [Bibtex]

Acknowledgements

This template was originally made by Phillip Isola and Richard Zhang for a colorful ECCV project; the code can be found here.
In this work, we would like to thank Yiluan Guo (Motional), Jiamei Sun (Mircosoft) and Milad Abdollahzadeh (SUTD) for their fruitful discussion, comments and feedback.