Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
97 tokens/sec
GPT-4o
53 tokens/sec
Gemini 2.5 Pro Pro
43 tokens/sec
o3 Pro
4 tokens/sec
GPT-4.1 Pro
47 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Parameter Exchange for Robust Dynamic Domain Generalization (2311.13928v1)

Published 23 Nov 2023 in cs.CV and cs.AI

Abstract: Agnostic domain shift is the main reason of model degradation on the unknown target domains, which brings an urgent need to develop Domain Generalization (DG). Recent advances at DG use dynamic networks to achieve training-free adaptation on the unknown target domains, termed Dynamic Domain Generalization (DDG), which compensates for the lack of self-adaptability in static models with fixed weights. The parameters of dynamic networks can be decoupled into a static and a dynamic component, which are designed to learn domain-invariant and domain-specific features, respectively. Based on the existing arts, in this work, we try to push the limits of DDG by disentangling the static and dynamic components more thoroughly from an optimization perspective. Our main consideration is that we can enable the static component to learn domain-invariant features more comprehensively by augmenting the domain-specific information. As a result, the more comprehensive domain-invariant features learned by the static component can then enforce the dynamic component to focus more on learning adaptive domain-specific features. To this end, we propose a simple yet effective Parameter Exchange (PE) method to perturb the combination between the static and dynamic components. We optimize the model using the gradients from both the perturbed and non-perturbed feed-forward jointly to implicitly achieve the aforementioned disentanglement. In this way, the two components can be optimized in a mutually-beneficial manner, which can resist the agnostic domain shifts and improve the self-adaptability on the unknown target domain. Extensive experiments show that PE can be easily plugged into existing dynamic networks to improve their generalization ability without bells and whistles.

Definition Search Book Streamline Icon: https://streamlinehq.com
References (58)
  1. MetaReg: Towards Domain Generalization using Meta-Regularization. In NeurIPS, Vol. 31.
  2. Recognition in terra incognita. In ECCV. 456–473.
  3. Exploiting domain-specific features to enhance domain generalization. NeurlPS 34 (2021), 21189–21201.
  4. Domain generalization by solving jigsaw puzzles. In CVPR. 2229–2238.
  5. Swad: Domain generalization by seeking flat minima. NeurIPS 34 (2021), 22405–22418.
  6. Learning to balance specificity and invariance for in and out of domain generalization. In ECCV. 301–318.
  7. Compound Domain Generalization via Meta-Knowledge Encoding. In CVPR. 7119–7129.
  8. Self-Supervised Noisy Label Learning for Source-Free Unsupervised Domain Adaptation. In IROS. 10185–10192.
  9. A style and semantic memory mechanism for domain generalization. In ICCV. 9164–9173.
  10. Zhengming Ding and Yun Fu. 2017. Deep domain generalization with structured low-rank constraint. TIP 27, 1 (2017), 304–313.
  11. Domain generalization via model-agnostic learning of semantic features. NeurIPS 32 (2019).
  12. Learning to learn with variational information bottleneck for domain generalization. In ECCV. 200–216.
  13. Antonio D’Innocente and Barbara Caputo. 2018. Domain generalization with domain-specific aggregation modules. In GCPR. 187–198.
  14. The pascal visual object classes (voc) challenge. IJCV 88 (2010), 303–338.
  15. Unbiased metric learning: On the utilization of multiple datasets and web images for softening bias. In ICCV. 1657–1664.
  16. Learning generative visual models from few training examples: An incremental bayesian approach tested on 101 object categories. In CVPR. 178–178.
  17. Ishaan Gulrajani and David Lopez-Paz. 2021. In Search of Lost Domain Generalization. In ICLR.
  18. Deep Residual Learning for Image Recognition. In CVPR. 770–778.
  19. Domain and Content Adaptive Convolution based Multi-Source Domain Generalization for Medical Image Segmentation. TMI 42, 1 (2022), 233–244.
  20. Self-challenging improves cross-domain generalization. In ECCV. 124–140.
  21. Yusuke Iwasawa and Yutaka Matsuo. 2021. Test-time classifier adjustment module for model-agnostic domain generalization. NeurIPS 34 (2021), 2427–2440.
  22. Averaging weights leads to wider optima and better generalization. In UAI. 876–885.
  23. Selfreg: Self-supervised contrastive regularization for domain generalization. In ICCV. 9619–9628.
  24. Omni-Dimensional Dynamic Convolution. In ICLR.
  25. Deeper, broader and artier domain generalization. In ICCV. 5542–5550.
  26. Learning to generalize: Meta-learning for domain generalization. In AAAI, Vol. 32.
  27. Episodic training for domain generalization. In ICCV. 1446–1455.
  28. Domain generalization with adversarial feature learning. In CVPR. 5400–5409.
  29. Domain generalization for medical imaging classification with linear-dependency regularization. NeurIPS 33 (2020), 3118–3129.
  30. Uncertainty Modeling for Out-of-Distribution Generalization. In ICLR.
  31. Deep domain generalization via conditional invariant adversarial networks. In ECCV. 624–639.
  32. Dynamic Transfer for Multi-Source Domain Adaptation. In CVPR. 10998–11007.
  33. Attribute-Aware Convolutional Neural Networks for Facial Beauty Prediction.. In IJCAI. 847–853.
  34. Semi-Supervised Domain Generalization with Evolving Intermediate Domain. arXiv preprint arXiv:2111.10221 (2021).
  35. Toshihiko Matsuura and Tatsuya Harada. 2020. Domain generalization using a mixture of multiple latent domains. In AAAI, Vol. 34. 11749–11756.
  36. Attention diversification for domain generalization. In ECCV. 322–340.
  37. Unified deep supervised domain adaptation and generalization. In ICCV. 5715–5725.
  38. Reducing Domain Gap by Reducing Style Bias. In CVPR. 8690–8699.
  39. Moment matching for multi-source domain adaptation. In ICCV. 1406–1415.
  40. Efficient domain generalization via common-specific low-rank decomposition. In ICML. 7728–7738.
  41. LabelMe: a database and web-based tool for image annotation. IJCV 77 (2008), 157–173.
  42. Domain adversarial neural networks for domain generalization: When it works and how to improve. Machine Learning (2023), 1–37.
  43. Test-time training with self-supervision for generalization under distribution shifts. In ICML. 9229–9248.
  44. Dynamic Domain Generalization. In IJCAI. 1342–1348.
  45. Laurens Van der Maaten and Geoffrey Hinton. 2008. Visualizing data using t-SNE. JMLR 9, 11 (2008).
  46. Deep hashing network for unsupervised domain adaptation. In CVPR. 5018–5027.
  47. Hao Wang and Xiaojun Bi. 2022. Domain generalization and adaptation based on second-order style information. PR 127 (2022), 108595.
  48. Respecting domain relations: Hypothesis invariance for domain generalization. In ICPR. 9756–9763.
  49. Sun database: Large-scale scene recognition from abbey to zoo. In CVPR.
  50. Adapt then Generalize: A Simple Two-Stage Framework for Semi-Supervised Domain Generalization. In ICME.
  51. A fourier-based framework for domain generalization. In CVPR. 14383–14392.
  52. PCL: Proxy-based Contrastive Learning for Domain Generalization. In CVPR. 7097–7107.
  53. Mixup: Beyond Empirical Risk Minimization. In ICLR.
  54. Test-time Fourier Style Calibration for Domain Generalization. In IJCAI. 1721–1727.
  55. Learning to generalize unseen domains via memory-based multi-source meta-learning for person re-identification. In CVPR. 6277–6286.
  56. Deep domain-adversarial image generation for domain generalisation. In AAAI, Vol. 34. 13025–13032.
  57. Learning to generate novel domains for domain generalization. In ECCV. 561–578.
  58. Domain Generalization with MixStyle. In ICLR.
User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (6)
  1. Luojun Lin (13 papers)
  2. Zhifeng Shen (5 papers)
  3. Zhishu Sun (3 papers)
  4. Yuanlong Yu (17 papers)
  5. Lei Zhang (1691 papers)
  6. Weijie Chen (52 papers)
Citations (5)
View on Semantic Scholar

Summary

We haven't generated a summary for this paper yet.

Ai Sparkles Streamline Icon: https://streamlinehq.com Summarize Now