Attracting and Dispersing: A Simple Approach for Source-free Domain Adaptation

Authors: Shiqi Yang, yaxing wang, kai wang, Shangling Jui, Joost van de Weijer

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The experimental results prove the superiority of our method, and our method can be adopted as a simple but strong baseline for future research in SFDA. Our method can be also adapted to source-free open-set and partial-set DA which further shows the generalization ability of our method. Code is available in https://github.com/Albert0147/Aa D_SFDA.
Researcher Affiliation Collaboration Shiqi Yang1, Yaxing Wang2 , Kai Wang1, Shangling Jui3, Joost van de Weijer1 1 Computer Vision Center, Universitat Autonoma de Barcelona, Barcelona, Spain 2 Nankai University, Tianjin, China 3 Huawei Kirin Solution, Shanghai, China
Pseudocode Yes Algorithm 1 Attracting and Dispersing for SFDA Require: Source-pretrained model and target data Dt 1: Build memory bank storing all target features and predictions 2: while Adaptation do 3: Sample batch T from Dt and Update memory bank 4: For each feature zi in T , retrieve K-nearest neighbors (Ci) and their predictions from memory bank 5: Update model by minimizing Eq. 5 6: end while
Open Source Code Yes Code is available in https://github.com/Albert0147/Aa D_SFDA.
Open Datasets Yes We conduct experiments on three benchmark datasets for image classification: Office-31, Office-Home and Vis DA-C 2017.
Dataset Splits No The paper does not explicitly provide training/test/validation dataset splits, such as specific percentages or counts for validation data.
Hardware Specification No The paper states: 'Did you include the total amount of compute and the type of resources used (e.g., type of GPUs, internal cluster, or cloud provider)? [No]'
Software Dependencies No The paper does not provide specific version numbers for software dependencies or libraries used (e.g., PyTorch version, CUDA version).
Experiment Setup Yes We adopt SGD with momentum 0.9 and batch size of 64 for all datasets. The learning rate for Office-31 and Office-Home is set to 1e-3 for all layers, except for the last two newly added fc layers, where we apply 1e-2. Learning rates are set 10 times smaller for Vis DA. We train 40 epochs for Office-31 and Office-Home while 15 epochs for Vis DA.