2022
DOI: 10.48550/arxiv.2204.13263
|View full text |Cite
Preprint
|
Sign up to set email alerts
|

Covariance-aware Feature Alignment with Pre-computed Source Statistics for Test-time Adaptation

Abstract: The accuracy of deep neural networks is degraded when the distribution of features in the test environment (target domain) differs from that of the training (source) environment. To mitigate the degradation, test-time adaptation (TTA), where a model adapts to the target domain without access to the source dataset, can be used in the test environment. However, the existing TTA methods lack feature distribution alignment between the source and target domains, which unsupervised domain adaptation mainly addresses… Show more

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
1
1
1

Citation Types

0
3
0

Year Published

2023
2023
2023
2023

Publication Types

Select...
1

Relationship

0
1

Authors

Journals

citations
Cited by 1 publication
(3 citation statements)
references
References 17 publications
0
3
0
Order By: Relevance
“…Test-time adaptation allows the model to adapt to the test data (i.e., target domain) in a source-free and online manner [33,61,64]. Existing works improve TTA performance with sophisticated designs of unsupervised loss [48,71,42,9,45,57,5,16,1,3,12] or enhance the usability of small batch sizes [36,69,31,51,40] considering streaming test data. They focus on improving the adaptation performance with a stationary target domain (i.e., single domain TTA setup).…”
Section: Related Workmentioning
confidence: 99%
See 2 more Smart Citations
“…Test-time adaptation allows the model to adapt to the test data (i.e., target domain) in a source-free and online manner [33,61,64]. Existing works improve TTA performance with sophisticated designs of unsupervised loss [48,71,42,9,45,57,5,16,1,3,12] or enhance the usability of small batch sizes [36,69,31,51,40] considering streaming test data. They focus on improving the adaptation performance with a stationary target domain (i.e., single domain TTA setup).…”
Section: Related Workmentioning
confidence: 99%
“…Convolution layer in meta networks. As the hyperparameters of the convolution layer 1 , we set the bias to false and Table 15. Ablation of the combination of transformations.…”
Section: Further Implementation Detailsmentioning
confidence: 99%
See 1 more Smart Citation