Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

TTA-DAME: Test-Time Adaptation with Domain Augmentation and Model Ensemble for Dynamic Driving Conditions

Created by
  • Haebom

Author

Dongjae Jeon, Taeheon Kim, Seongwon Cho, Minhyuk Seo, Jonghyun Choi

Outline

This paper presents a method for enabling models to dynamically adapt and perform optimally in changing target domains through Test-time Adaptation (TTA). Specifically, we focus on real-world driving environments with frequent weather changes, and propose a method called TTA-DAME. TTA-DAME applies source domain data augmentation to the target domain and introduces a domain discriminator and a specialized domain detector to mitigate abrupt domain changes, particularly from day to night. We train multiple detectors and integrate prediction results using Non-Maximum Suppression (NMS) to enhance adaptability. The effectiveness of the proposed method is experimentally verified using the SHIFT Benchmark.

Takeaways, Limitations

•
Takeaways:
◦
We present a TTA method that effectively responds to dynamic domain changes such as actual driving environments.
◦
Mitigate domain variation by augmenting source domain data and leveraging domain discriminators and detectors.
◦
Improved adaptability by integrating multi-detector prediction results through NMS.
◦
Experimentally verified performance improvements in SHIFT Benchmark.
•
Limitations:
◦
Further research is needed on the generalization performance of the proposed method.
◦
Verification of versatility across various domain change types is required.
◦
Additional experiments using benchmark datasets other than SHIFT Benchmark are needed.
◦
Research is needed to analyze computational costs and improve efficiency.
👍
Made with Slashpage