Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Domain Adaptation for Open-Vocabulary Segmentation
Open–vocabulary segmentation [1, 2] offers amazing zero-shot semantic perception capabilities that generalise to a large range of environments, but is less precise compared to models that are trained on the specific domain and sensor setup. We want to leverage unsupervised domain adaptation [3] to transfer the knowledge of open-vocabulary segmentation to a domain and sensor specific model.
[1] Ghiasi et al. Open-Vocabulary Image Segmentation, https://arxiv.org/abs/2112.12143
[2] Peng et al. OpenScene, 2022 https://arxiv.org/abs/2211.15654
[3] Liu et al. Unsupervised Continual Semantic Adaptation, https://arxiv.org/abs/2211.13969