yuejiangliu / awesome-source-free-test-time-adaptation Goto Github PK

Hi Yuejiang,

Thanks for maintaining this awesome list!

I would like to request to add our ECCV 2022 paper: "Source-free Video Domain Adaptation by Learning Temporal Consistency for Action Recognition" (URL: https://www.ecva.net/papers/eccv_2022/papers_ECCV/papers/136940144.pdf https://arxiv.org/abs/2203.04559) (Project URL: https://xuyu0010.github.io/sfvda.html). This is one of the pioneer works in source-free video domain adaptation (SFVDA) where we tackle the task by learning temporal consistency across local temporal features.

I would also like to request to add our Preprint paper: "Multi-Modal Continual Test-Time Adaptation for 3D Semantic Segmentation" (URL: https://arxiv.org/abs/2303.10457) where we explore Multi-Modal Continual Test-Time Adaptation (MM-CTTA) as a new extension of Continual Test-Time Adaptation (CTTA) for 3D semantic segmentation by adaptively attend to the reliable modality while avoiding catastrophic forgetting during continual domain shifts.

Thank you very much!
Yuecong

Hi Yuejiang,

Thanks for maintaining the list! We would like to request to add our NeurIPS2022 paper: "Meta-DMoE: Adapting to Domain Shift by Meta-Distillation from Mixture-of-Experts". We formulate the source free test time adaptation as the knowledge distillation process to solve the domain shift problem. At test-time, the model is adapted to a target domain using unlabeled data in an unsupervised manner.

Thanks.

Hello Yuejiang Liu,
Thanks for compiling this nice list.
I just wanted to point out our 2020 MICCAI work: Source Free Domain Adaptation for Image Segmentation, which could be a nice addition to your list.
https://arxiv.org/abs/2108.03152
Cheers !
Mathilde

Hi

My work [1] on ICLR 2023 proposes a new test time adaptation layer called ShiftMatch that matches the second order statistics of the test features to that of the training features for better robustness, I am wondering if it can fit into the feature alignment or the batch normalization section in this list.

[1] https://openreview.net/forum?id=kUI41mY8bHl

Thanks,
Xi

Hi Yuejiang,

Thanks for maintaining this awesome list!

I would like to request to add our paper: Feature Alignment and Uniformity for Test Time Adaptation (CVPR 2023).
arxiv: https://arxiv.org/abs/2303.10902
code: https://github.com/SakurajimaMaiii/TSD
This belongs to online test time adaptation where the model can only access online unlabeled test samples and pre-trained models on the training domains. We propose two novle loss functions in a feature alignment & uniformity perspective.

Thanks for considering this.
Shuai Wang

Hi Yuejiang,

Thanks for maintaining this awesome list!

I would like to request to add our paper: "On the Robustness of Open-World Test-Time Training: Self-Training with Dynamic Prototype Expansion" (ICCV 2023 Oral).

ArXiv: https://arxiv.org/abs/2308.09942
Code: https://github.com/Yushu-Li/OWTTT
Project page: https://yushu-li.github.io/owttt-site/

In this work, we consider the problem of test-time adaptation in open-world scenarios for the first time. Furthermore, we used the self-training method based on prototype expansion to achieve SOTA performance under the proposed protocol. I think this paper belongs to both "Class Prototype", "Self-supervision", and a new category "Open World".

Thanks for considering this.
Yushu Li