Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
156 tokens/sec
GPT-4o
7 tokens/sec
Gemini 2.5 Pro Pro
45 tokens/sec
o3 Pro
4 tokens/sec
GPT-4.1 Pro
38 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Towards Improved Proxy-based Deep Metric Learning via Data-Augmented Domain Adaptation (2401.00617v1)

Published 1 Jan 2024 in cs.CV

Abstract: Deep Metric Learning (DML) plays an important role in modern computer vision research, where we learn a distance metric for a set of image representations. Recent DML techniques utilize the proxy to interact with the corresponding image samples in the embedding space. However, existing proxy-based DML methods focus on learning individual proxy-to-sample distance while the overall distribution of samples and proxies lacks attention. In this paper, we present a novel proxy-based DML framework that focuses on aligning the sample and proxy distributions to improve the efficiency of proxy-based DML losses. Specifically, we propose the Data-Augmented Domain Adaptation (DADA) method to adapt the domain gap between the group of samples and proxies. To the best of our knowledge, we are the first to leverage domain adaptation to boost the performance of proxy-based DML. We show that our method can be easily plugged into existing proxy-based DML losses. Our experiments on benchmarks, including the popular CUB-200-2011, CARS196, Stanford Online Products, and In-Shop Clothes Retrieval, show that our learning algorithm significantly improves the existing proxy losses and achieves superior results compared to the existing methods.

Definition Search Book Streamline Icon: https://streamlinehq.com
References (82)
  1. A theory of learning from different domains. Machine learning, 79: 151–175.
  2. A unifying mutual information view of metric learning: cross-entropy vs. pairwise losses. In ECCV, 548–564. Springer.
  3. Mode regularized generative adversarial networks. ICLR.
  4. Reusing the Task-specific Classifier as a Discriminator: Discriminator-free Adversarial Domain Adaptation. In CVPR, 7181–7190.
  5. Person re-identification by multi-channel parts-based cnn with improved triplet loss function. In CVPR, 1335–1344.
  6. Learning a similarity metric discriminatively, with application to face verification. In CVPR, volume 1, 539–546. IEEE.
  7. Towards discriminability and diversity: Batch nuclear-norm maximization under label insufficient situations. In CVPR, 3941–3950.
  8. Batch dropblock network for person re-identification and beyond. In ICCV, 3691–3701.
  9. Imagenet: A large-scale hierarchical image database. In CVPR, 248–255. Ieee.
  10. Training vision transformers for image retrieval. arXiv preprint arXiv:2102.05644.
  11. Hyperbolic Vision Transformers: Combining Improvements in Metric Learning. In CVPR, 7409–7419.
  12. Unsupervised domain adaptation by backpropagation. In ICML, 1180–1189. PMLR.
  13. Domain-adversarial training of neural networks. The journal of machine learning research, 17(1): 2096–2030.
  14. Self-supervising fine-grained region similarities for large-scale image localization. In ECCV, 369–386. Springer.
  15. Neighbourhood components analysis. NIPS, 17.
  16. Generative adversarial networks. Communications of the ACM, 63(11): 139–144.
  17. Dimensionality reduction by learning an invariant mapping. In CVPR, volume 2, 1735–1742. IEEE.
  18. A Survey on Adversarial Domain Adaptation. Neural Processing Letters, 1–41.
  19. Deep residual learning for image recognition. In CVPR, 770–778.
  20. In defense of the triplet loss for person re-identification. arXiv preprint arXiv:1703.07737.
  21. Batch normalization: Accelerating deep network training by reducing internal covariate shift. In ICML, 448–456. PMLR.
  22. Image-to-image translation with conditional adversarial networks. In CVPR, 1125–1134.
  23. Coded Residual Transform for Generalizable Deep Metric Learning. NeurIPS.
  24. Not all samples are created equal: Deep learning with importance sampling. In ICML, 2525–2534. PMLR.
  25. Proxy anchor loss for deep metric learning. In CVPR, 3238–3247.
  26. Adam: A method for stochastic optimization. ICLR.
  27. Auto-encoding variational bayes. arXiv preprint arXiv:1312.6114.
  28. Learning with memory-based virtual classes for deep metric learning. In CVPR, 11792–11801.
  29. 3d object representations for fine-grained categorization. In ICCV workshop, 554–561.
  30. M-adda: Unsupervised domain adaptation with deep metric learning. Domain adaptation for visual understanding, 17–31.
  31. Sliced wasserstein discrepancy for unsupervised domain adaptation. In CVPR, 10285–10295.
  32. Rank-based distance metric learning: An application to image retrieval. In CVPR, 1–8. IEEE.
  33. Hypergraph-Induced Semantic Tuplet Loss for Deep Metric Learning. In CVPR, 212–222.
  34. DAS: Densely-Anchored Sampling for Deep Metric Learning. ECCV.
  35. Deepfashion: Powering robust clothes recognition and retrieval with rich annotations. In CVPR, 1096–1104.
  36. Localize me anywhere, anytime: a multi-task point-retrieval approach. In ICCV, 2434–2442.
  37. Diva: Diverse visual feature aggregation for deep metric learning. In ECCV, 590–607. Springer.
  38. No fuss distance metric learning using proxies. In CVPR, 360–368.
  39. A metric learning reality check. In ECCV, 681–699. Springer.
  40. Deep metric learning via facility location. In CVPR, 5382–5390.
  41. Deep metric learning via lifted structured feature embedding. In CVPR, 4004–4012.
  42. Recall@ k surrogate loss with large batches and similarity mixup. In CVPR, 7502–7511.
  43. Multi-adversarial domain adaptation. In AAAI.
  44. Pinheiro, P. O. 2018. Unsupervised domain adaptation with similarity learning. In CVPR, 8004–8013.
  45. Softtriple loss: Deep metric learning without triplet sampling. In ICCV, 6450–6458.
  46. Covariate shift and local learning by distribution matching.
  47. Unsupervised representation learning with deep convolutional generative adversarial networks. arXiv preprint arXiv:1511.06434.
  48. Improved image description via embedded object structure graph and semantic feature matching. In ISM, 73–80. IEEE.
  49. Beyond the deep metric learning: enhance the cross-modal matching with adversarial discriminative domain regularization. In ICPR, 10165–10172. IEEE.
  50. Improving diversity of image captioning through variational autoencoders and adversarial learning. In WACV, 263–272. IEEE.
  51. Metric learning with adaptive density discrimination. ICLR.
  52. Simultaneous similarity-based self-distillation for deep metric learning. In ICML, 9095–9106. PMLR.
  53. Revisiting training strategies and generalization performance in deep metric learning. In ICML, 8242–8252. PMLR.
  54. Non-isotropy Regularization for Proxy-based Deep Metric Learning. In CVPR, 7420–7430.
  55. Neighbourhood component analysis. NIPS, 17(513-520): 4.
  56. Maximum classifier discrepancy for unsupervised domain adaptation. In CVPR, 3723–3732.
  57. Facenet: A unified embedding for face recognition and clustering. In CVPR, 815–823.
  58. Learning intra-batch connections for deep metric learning. In ICML, 9410–9421. PMLR.
  59. Sohn, K. 2016. Improved deep metric learning with multi-class n-pair loss objective. NIPS, 29.
  60. Stochastic class-based hard example mining for deep metric learning. In CVPR, 7251–7259.
  61. Instance-level image retrieval using reranking transformers. In ICCV, 12105–12115.
  62. Proxynca++: Revisiting and revitalizing proxy neighborhood component analysis. In ECCV, 448–464. Springer.
  63. Unbiased look at dataset bias. In CVPR, 1521–1528. IEEE.
  64. Adversarial discriminative domain adaptation. In CVPR, 7167–7176.
  65. It takes two to tango: Mixup for deep metric learning. ICLR.
  66. The caltech-ucsd birds-200-2011 dataset.
  67. Adversarial cross-modal retrieval. In Multimedia, 154–162.
  68. Learning fine-grained image similarity with deep ranking. In CVPR, 1386–1393.
  69. Deep visual domain adaptation: A survey. Neurocomputing, 312: 135–153.
  70. Multi-similarity loss with general pair weighting for deep metric learning. In CVPR, 5022–5030.
  71. Cross-batch memory for embedding learning. In CVPR, 6388–6397.
  72. Distance metric learning for large margin nearest neighbor classification. Journal of machine learning research, 10(2).
  73. Deep cosine metric learning for person re-identification. In WACV, 748–756. IEEE.
  74. Sampling matters in deep embedding learning. In CVPR, 2840–2848.
  75. Hard negative examples are hard, but useful. In ECCV, 126–142. Springer.
  76. Retrieving and classifying affective images via deep metric learning. In AAAI, volume 32.
  77. Hierarchical proxy-based loss for deep metric learning. In WACV, 1859–1868.
  78. Deep metric learning for person re-identification. In ICPR, 34–39. IEEE.
  79. Attributable Visual Similarity Learning. In CVPR, 7532–7541.
  80. Deep compositional metric learning. In CVPR, 9320–9329.
  81. Deep relational metric learning. In ICCV, 12065–12074.
  82. Fewer is more: A deep graph metric learning perspective using fewer proxies. NeurIPS, 33: 17792–17803.
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