Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Open-World Test-Time Training: Self-Training with Contrast Learning

Houcheng Su, Mengzhu Wang, Jiao Li, Bingli Wang, Daixian Liu, Zeheng Wang

TL;DR

This work introduces Open World Dynamic Contrastive Learning (OWDCL), an innovative approach that utilizes contrastive learning to augment positive sample pairs and significantly enhances model robustness in subsequent stages of test-time training.

Abstract

Traditional test-time training (TTT) methods, while addressing domain shifts, often assume a consistent class set, limiting their applicability in real-world scenarios characterized by infinite variety. Open-World Test-Time Training (OWTTT) addresses the challenge of generalizing deep learning models to unknown target domain distributions, especially in the presence of strong Out-of-Distribution (OOD) data. Existing TTT methods often struggle to maintain performance when confronted with strong OOD data. In OWTTT, the focus has predominantly been on distinguishing between overall strong and weak OOD data. However, during the early stages of TTT, initial feature extraction is hampered by interference from strong OOD and corruptions, resulting in diminished contrast and premature classification of certain classes as strong OOD. To address this, we introduce Open World Dynamic Contrastive Learning (OWDCL), an innovative approach that utilizes contrastive learning to augment positive sample pairs. This strategy not only bolsters contrast in the early stages but also significantly enhances model robustness in subsequent stages. In comparison datasets, our OWDCL model has produced the most advanced performance.

Open-World Test-Time Training: Self-Training with Contrast Learning

TL;DR

This work introduces Open World Dynamic Contrastive Learning (OWDCL), an innovative approach that utilizes contrastive learning to augment positive sample pairs and significantly enhances model robustness in subsequent stages of test-time training.

Abstract

Traditional test-time training (TTT) methods, while addressing domain shifts, often assume a consistent class set, limiting their applicability in real-world scenarios characterized by infinite variety. Open-World Test-Time Training (OWTTT) addresses the challenge of generalizing deep learning models to unknown target domain distributions, especially in the presence of strong Out-of-Distribution (OOD) data. Existing TTT methods often struggle to maintain performance when confronted with strong OOD data. In OWTTT, the focus has predominantly been on distinguishing between overall strong and weak OOD data. However, during the early stages of TTT, initial feature extraction is hampered by interference from strong OOD and corruptions, resulting in diminished contrast and premature classification of certain classes as strong OOD. To address this, we introduce Open World Dynamic Contrastive Learning (OWDCL), an innovative approach that utilizes contrastive learning to augment positive sample pairs. This strategy not only bolsters contrast in the early stages but also significantly enhances model robustness in subsequent stages. In comparison datasets, our OWDCL model has produced the most advanced performance.
Paper Structure (19 sections, 21 equations, 4 figures, 5 tables)

This paper contains 19 sections, 21 equations, 4 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Unsupervised Domain Adaptation
  4. Test-Time Training
  5. Open-Set Domain Adaptation
  6. Methods
  7. Problem Formulation
  8. Overall Test-Time Training Framework
  9. Contrastive Alignment by Positive Sample Pairs
  10. Contrastive Alignment by Cluster and Sample Pairs
  11. Experiments
  12. Datasets and Evaluation Metric
  13. Comparison Methods and Settings
  14. Comparative experiments
  15. Further Performance Analysis
  16. ...and 4 more sections

Figures (4)

  • Figure 1: In an experimental setup involving 15 types of corruption within the ImageNet-C dataset and employing the MNIST dataset as a benchmark for Strong OOD analysis, we conduct a performance comparison between OWDCL and OWTTT.
  • Figure 2: Overall framework of our model OWDCL. (1) $\mathcal{L}_{ps}$: Improve the feature extraction ability of the model by comparing samples with enhanced samples. (2)$\mathcal{L}_{cs}$: The classification accuracy is optimized through the comprehensive comparison between the enhanced sample pair and the class center of gravity.
  • Figure 3: Visual analysis experiment. Black is strong OOD, while the others are weak OOD.
  • Figure 4: Parameter Robustness Analysis.