ImDeeplabV3plus with instance selective whitening loss in domain generalization semantic segmentation

Semantic segmentation is a classical problem in computer vision, which important the field of autonomous driving. Although significant progress has been achieved semantic segmentation, its generalization ability to unknown domains still challenging. To effectively solve this problem, method ImDeeplabV3plus with instance selective whitening loss proposed paper. DeeplabV3plus selected as baseline. In order enhance representation region interest, coordinate attention (CA) mechanism added. better integrate multiple low-level features, adaptively spatial feature fusion (ASFF) employed learn importance features at different levels for each location. For preferably coping domain changes, an (ISW) introduced early stage backbone. The model trained Cityscapes dataset and then applied RobotCar dataset. Compared DeeplabV3plus, authors’ shows 1.29% mIoU improvement. When ISW added, 2.08% improvement compared ImDeeplabV3plus. Experimental results show that simple improves ability.