Learning to adapt structured output space for semantic
Download as PPTX, PDF0 likes157 views
This document discusses a method for adapting semantic segmentation models to new domains without labeled target data. It proposes a multi-level adversarial learning approach that transfers knowledge across domains by matching feature and output distributions between source and target data at multiple network levels. This helps generalize models to new, unlabeled images and close the domain gap to improve segmentation performance on different datasets.
1 of 10
Download to read offline
More Related Content
More from 哲东 郑 (20)
Ad
Recently uploaded (20)
Ad
Learning to adapt structured output space for semantic
- 1. Learning to Adapt Structured Output Space for Semantic Segmentation
- 2. Motivation • This trained model may not generalize well to unseen images, especially when there is a domain gap between the training (source) and test (target) images. • To address this issue, knowledge transfer or domain adaptation techniques have been proposed to close the gap between source and target domains, where annotations are not available in the target domain.