Partial Domain Adaptation on Semantic Segmentation

Article Engineering, Electrical & Electronic

Cross-Domain Complementary Learning Using Pose for Multi-Person Part Segmentation

Kevin Lin et al.

Summary: By utilizing skeleton representation, the proposed method effectively bridges the domain gap between synthetic and real data, enabling multi-person part segmentation without human-annotated labels. Leveraging the realistic variations in real data and obtainable labels in synthetic data, the method achieves comparable performance to state-of-the-art approaches without human labeling, and outperforms supervised methods significantly when part labels are available in real images during training.

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY (2021)

Add to Collection

Proceedings Paper Computer Science, Artificial Intelligence

Multi-Source Domain Adaptation with Collaborative Learning for Semantic Segmentation

Jianzhong He et al.

Summary: This paper proposes a novel multi-source domain adaptation framework based on collaborative learning for semantic segmentation. By aligning pixel value distribution and utilizing collaborative learning with unlabeled target domain data, the proposed method achieves significant improvement in domain adaptation performance.

2021 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION, CVPR 2021 (2021)

Add to Collection

Article Engineering, Electrical & Electronic

Rethinking Triplet Loss for Domain Adaptation

Weijian Deng et al.

Summary: This study focuses on domain adaptation, proposing a similarity guided constraint (SGC) method integrated into the network as a triplet loss to achieve co-location of source and target domain features at the class level. The method ensures cross-domain similarities are preserved, leading to intra-class compactness and inter-class separability.

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY (2021)

Add to Collection

Article Computer Science, Artificial Intelligence