A Fully Convolutional Tri-branch Network (FCTN) for Domain Adaptation
This work addresses domain adaptation in urban scene segmentation, which is crucial for applications like autonomous driving, but it appears incremental as it builds on existing pseudo-labeling and multi-branch network approaches.
The paper tackles domain adaptation for urban scene segmentation by proposing a fully convolutional tri-branch network that alternates pseudo-labeling and re-training to learn target-specific representations, achieving state-of-the-art performance with significant improvements over previous methods on synthetic and real image datasets.
A domain adaptation method for urban scene segmentation is proposed in this work. We develop a fully convolutional tri-branch network, where two branches assign pseudo labels to images in the unlabeled target domain while the third branch is trained with supervision based on images in the pseudo-labeled target domain. The re-labeling and re-training processes alternate. With this design, the tri-branch network learns target-specific discriminative representations progressively and, as a result, the cross-domain capability of the segmenter improves. We evaluate the proposed network on large-scale domain adaptation experiments using both synthetic (GTA) and real (Cityscapes) images. It is shown that our solution achieves the state-of-the-art performance and it outperforms previous methods by a significant margin.