Unsupervised detection of contrast enhanced highlight landmarks

Article Computer Science, Artificial Intelligence

From Keypoints to Object Landmarks via Self-Training Correspondence: A Novel Approach to Unsupervised Landmark Discovery

Dimitrios Mallis et al.

Summary: This paper proposes a novel paradigm for unsupervised learning of object landmark detectors. A self-training approach is proposed where a landmark detector and descriptor improve themselves by training on generic keypoints. An iterative algorithm is used to alternate between producing new pseudo-labels through feature clustering and learning distinctive features for each pseudo-class through contrastive learning. Experimental results show that the proposed method achieves new state of the art results on various difficult datasets.

IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE (2023)

添加到收藏夹

Article Computer Science, Information Systems

Self-Supervised Masking for Unsupervised Anomaly Detection and Localization

Chaoqin Huang et al.

Summary: In this research, a self-supervised learning approach named Self-Supervised Masking (SSM) is proposed for unsupervised anomaly detection and localization. It utilizes random masking and restoration of images for training, and introduces a progressive mask refinement method to improve the efficiency and effectiveness of anomaly detection and localization.

IEEE TRANSACTIONS ON MULTIMEDIA (2023)

添加到收藏夹

Article Computer Science, Artificial Intelligence

BRULE: Barycenter-Regularized Unsupervised Landmark Extraction

Iaroslav Bespalov et al.

Summary: This work proposes a new unsupervised approach to detect landmarks in images by auto-encoding the wanted landmarks in the latent space. The interpretable latent space representation is achieved through a two-step regularization approach. Experimental results in unsupervised and semi-supervised training scenarios demonstrate the effectiveness of the proposed method, showing improved detection quality beyond state-of-the-art face models.

PATTERN RECOGNITION (2022)

添加到收藏夹

Proceedings Paper Computer Science, Artificial Intelligence

TransPose: Keypoint Localization via Transformer

Sen Yang et al.

Summary: The TransPose model introduces Transformer for human pose estimation, efficiently capturing long-range relationships and revealing dependencies of keypoints. The heatmap-based approach provides fine-grained image-specific dependencies, showing evidence of how the model handles special cases such as occlusion.

2021 IEEE/CVF INTERNATIONAL CONFERENCE ON COMPUTER VISION (ICCV 2021) (2021)

添加到收藏夹

Proceedings Paper Computer Science, Artificial Intelligence