We propose a test-time adaptation method for cross-domain image segmentation. Our method is simple: Given a new unseen instance at test time, we adapt a pre-trained model by conducting instance-specific BatchNorm (statistics) calibration. Our approach has two core components. First, we replace the manually designed BatchNorm calibration rule with a learnable module. Second, we leverage strong data augmentation to simulate random domain shifts for learning the calibration rule. In contrast to existing domain adaptation methods, our method does not require accessing the target domain data at training time or conducting computationally expensive test-time model training/optimization. Equipping our method with models trained by standard recipes achieves significant improvement, comparing favorably with several state-of-the-art domain generalization and one-shot unsupervised domain adaptation approaches. Combining our method with the domain generalization methods further improves performance, reaching a new state of the art.

(a) At training time, we learn the BatchNorm calibration rule by training only the newly initialized parameters on the strongly-augmented source domain data; (b) At test time, we conduct instance-specific BatchNorm calibration using the learned calibration rule. Note that our method does not perform test-time training or optimization, and thus the model parameters are fixed after training.

@inproceedings{zou2022learning,
  title={Learning Instance-Specific Adaptation for Cross-Domain Segmentation},
  author={Zou, Yuliang and Zhang, Zizhao and Li, Chun-Liang and Zhang, Han and Pfister, Tomas and Huang, Jia-Bin},
  booktitle={European Conference on Computer Vision},
  year={2022}
}