mlsanityData to 14 Oct 2026

How to Fine-tune Models with Few Samples: Update, Data Augmentation, and Test-time Augmentation

arXiv 2205.07874 ↗First posted 13 May 2022cs.LG cs.CVcited by 6

Not on the map. Why?

Abstract (arXiv)

Most of the recent few-shot learning (FSL) algorithms are based on transfer learning, where a model is pre-trained using a large amount of source data, and the pre-trained model is fine-tuned using a small amount of target data. In transfer learning-based FSL, sophisticated pre-training methods have been widely studied for universal representation. Therefore, it has become more important to utilize the universal representation for downstream tasks, but there are few studies on fine-tuning in FSL. In this paper, we focus on how to transfer pre-trained models to few-shot downstream tasks from the three perspectives: update, data augmentation, and test-time augmentation. First, we compare the two popular update methods, full fine-tuning (i.e., updating the entire network, FT) and linear probing (i.e., updating only a linear classifier, LP). We find that LP is better than FT with extremely few samples, whereas FT outperforms LP as training samples increase. Next, we show that data augmentation cannot guarantee few-shot performance improvement and investigate the effectiveness of data augmentation based on the intensity of augmentation. Finally, we adopt augmentation to both a support set for update (i.e., data augmentation) as well as a query set for prediction (i.e., test-time augmentation), considering support-query distribution shifts, and improve few-shot performance. The code is available at https://github.com/kimyuji/updating_FSL.

Key facts. Cited by 6 papers in the map (data through 14 Oct 2026). Builds on 49 of its 78 references that are in the map. First posted on arXiv 13 May 2022.

Builds on

49 of 78 references.

References · 49

  1. 2015Deep Residual Learning for Image Recognitioncited by 52,096
  2. 2019PyTorch: An Imperative Style, High-Performance Deep Learning Librarycited by 20,305
  3. 2016Densely Connected Convolutional Networkscited by 6,720
  4. 2019EfficientNet: Rethinking Model Scaling for Convolutional Neural Networkscited by 6,442
  5. 2017mixup: Beyond Empirical Risk Minimizationcited by 5,183
  6. 2017Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networkscited by 4,897
  7. 2017Prototypical Networks for Few-shot Learningcited by 3,015
  8. 2019CutMix: Regularization Strategy to Train Strong Classifiers with Localizable Featurescited by 2,884
  9. 2016Matching Networks for One Shot Learningcited by 2,255
  10. 2014Conditional Generative Adversarial Netscited by 2,162
  11. 2017EuroSAT: A Novel Dataset and Deep Learning Benchmark for Land Use and Land Cover Classificationcited by 1,496
  12. 2017Learning to Compare: Relation Network for Few-Shot Learningcited by 1,143
37 more
  1. 2022Fine-Tuning can Distort Pretrained Features and Underperform Out-of-Distributioncited by 739
  2. 2017Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networkscited by 707
  3. 2019A Closer Look at Few-shot Classificationcited by 684
  4. 2019Skin Lesion Analysis Toward Melanoma Detection 2018: A Challenge Hosted by the International Skin Imaging Collaboration (ISIC)cited by 669
  5. 2018Do Better ImageNet Models Transfer Better?cited by 565
  6. 2019Meta-Learning with Implicit Gradientscited by 525
  7. 2018Meta-Learning for Semi-Supervised Few-Shot Classificationcited by 456
  8. 2019A Large-scale Study of Representation Learning with the Visual Task Adaptation Benchmarkcited by 425
  9. 2020Fine-Tuning Pretrained Language Models: Weight Initializations, Data Orders, and Early Stoppingcited by 384
  10. 2017Meta-SGD: Learning to Learn Quickly for Few-Shot Learningcited by 379
  11. 2017The Effectiveness of Data Augmentation in Image Classification using Deep Learningcited by 370
  12. 2019A Baseline for Few-Shot Image Classificationcited by 298
  13. 2017Data Augmentation Generative Adversarial Networkscited by 278
  14. 2020Puzzle Mix: Exploiting Saliency and Local Statistics for Optimal Mixupcited by 261
  15. 2016Using Deep Learning for Image-Based Plant Disease Detectioncited by 219
  16. 2020How Well Do Self-Supervised Models Transfer?cited by 213
  17. 2019SimpleShot: Revisiting Nearest-Neighbor Classification for Few-Shot Learningcited by 195
  18. 2020Cross-Domain Few-Shot Classification via Learned Feature-Wise Transformationcited by 185
  19. 2018Understanding Measures of Uncertainty for Adversarial Example Detectioncited by 177
  20. 2018Large Scale Fine-Grained Categorization and Domain-Specific Transfer Learningcited by 174
  21. 2021Free Lunch for Few-shot Learning: Distribution Calibrationcited by 165
  22. 2019Charting the Right Manifold: Manifold Mixup for Few-shot Learningcited by 136
  23. 2020Embedding Propagation: Smoother Manifold for Few-Shot Classificationcited by 80
  24. 2020Self-training for Few-shot Transfer Across Extreme Task Differencescited by 76
  25. 2021Partial Is Better Than All: Revisiting Fine-tuning Strategy for Few-shot Learningcited by 72
  26. 2018Deflecting Adversarial Attacks with Pixel Deflectioncited by 71
  27. 2020Rethinking the Hyperparameters for Fine-tuningcited by 69
  28. 2020Learning Loss for Test-Time Augmentationcited by 56
  29. 2021Boosting the Generalization Capability in Cross-Domain Few-shot Learning via Noise-enhanced Supervised Autoencodercited by 51
  30. 2021Dynamic Distillation Network for Cross-Domain Few-Shot Recognition with Unlabeled Datacited by 50
  31. 2017Label Efficient Learning of Transferable Representations across Domains and Taskscited by 49
  32. 2020Affinity and Diversity: Quantifying Mechanisms of Data Augmentationcited by 45
  33. 2020A Closer Look at Few-Shot Crosslingual Transfer: The Choice of Shots Matterscited by 45
  34. 2021A Closer Look at How Fine-tuning Changes BERTcited by 42
  35. 2020Data Augmentation for Meta-Learningcited by 38
  36. 2017Image Classification of Melanoma, Nevus and Seborrheic Keratosis by Deep Neural Network Ensemblecited by 23
  37. 2022GDC- Generalized Distribution Calibration for Few-Shot Learningcited by 5

Built on by

Cited by year

Citing papers per year: 2023: 1; 2024: 3; 2025: 2.

6 papers.

  1. 2023A Billion-scale Foundation Model for Remote Sensing Imagescited by 62
  2. 2025Few-Shot Segmentation of Historical Maps via Linear Probing of Vision Foundation Modelscited by 3
  3. 2024Token-Efficient Leverage Learning in Large Language Modelscited by 2
  4. 2024FSL-Rectifier: Rectify Outliers in Few-Shot Learning via Test-Time Augmentationcited by 1
  5. 2024Enhancing Fine-Grained Visual Recognition in the Low-Data Regime Through Feature Magnitude Regularizationcited by 0
  6. 2025Domain Borders Are There to Be Crossed With Federated Few-Shot Adaptationcited by 0