Proxy Anchor-based Unsupervised Learning for Continuous Generalized Category Discovery

motivation: understanding CGCD

self notes

Research

• Methodology::
• Key Contructs::
  • IVs::
  • DVs::
  • Moderators::
  • Others::
• Key_Findings::
• Contributions::
• Limitations::

Self Critique

• Critique
  • Strengths
  • Limitations
• How is it relevant to my research?
  • Relevant_topic::
  • Use::

[!Cite] [1] H. Kim, S. Suh, D. Kim, D. Jeong, H. Cho와/과J. Kim, “Proxy Anchor-based Unsupervised Learning for Continuous Generalized Category Discovery”. arXiv, 2023년 11월 2일. 접근된: 2024년 2월 13일. [Online]. Available at: http://arxiv.org/abs/2307.10943

[!Synth] Contribution::

[!Md]

Author:: Kim, Hyungmin>

Author:: Suh, Sungho>

Author:: Kim, Daehwan>

Author:: Jeong, Daun>

Author:: Cho, Hansang>

Author:: Kim, Junmo>

Title:: Proxy Anchor-based Unsupervised Learning for Continuous Generalized Category Discovery Year:: 2023 Citekey:: [[@2023-01-01-ProxyAnchorbasedUnsupervisedKimEtAl2023]]

Tags:: Computer Science - Artificial Intelligence, Computer Science - Computer Vision and Pattern Recognition itemType:: preprint

[!Link] Kim 등 - 2023 - Proxy Anchor-based Unsupervised Learning for Conti.pdf

[!Abstract]

abstract:: Recent advances in deep learning have significantly improved the performance of various computer vision applications. However, discovering novel categories in an incremental learning scenario remains a challenging problem due to the lack of prior knowledge about the number and nature of new categories. Existing methods for novel category discovery are limited by their reliance on labeled datasets and prior knowledge about the number of novel categories and the proportion of novel samples in the batch. To address the limitations and more accurately reflect real-world scenarios, in this paper, we propose a novel unsupervised class incremental learning approach for discovering novel categories on unlabeled sets without prior knowledge. The proposed method fine-tunes the feature extractor and proxy anchors on labeled sets, then splits samples into old and novel categories and clusters on the unlabeled dataset. Furthermore, the proxy anchors-based exemplar generates representative category vectors to mitigate catastrophic forgetting. Experimental results demonstrate that our proposed approach outperforms the state-of-the-art methods on finegrained datasets under real-world scenarios.

---

Annotations

1. Introduction

2. Problem Definition

2.1. Continuous Generalized Category Discovery

2.2. Setting of Continuous Generalized Category Discovery

3. Method

3.1. Initial Step: Fine Tune

3.2. Discovering Novel Categories Step

Separation:

Pseudo-labeling:

3.3. Category Incremental Step

Training modified model and PAs:

4. Experimental Results

4.1. Implementation Details

4.2. Evaluation Metrics

5. Related work

5.1. Novel Category Discovery

5.2. Image Retrieval

5.3. Noise Label

6. Conclusion

Acknowledgements

---