一种弱监督开放词汇物体检测方法、系统及设备
By combining adaptive masking and visual prototype alignment at the feature level with instance completion at the label level, the problems of incomplete target localization and low quality of pseudo-labels in weakly supervised visual target detection are solved, thereby improving the accuracy and robustness of the detection model.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HUNAN NORMAL UNIVERSITY
- Filing Date
- 2026-04-27
- Publication Date
- 2026-07-17
AI Technical Summary
Existing weakly supervised visual target detection technologies suffer from incomplete target localization, large open-vocabulary classification bias, and low-quality pseudo-labels, which limit detector performance.
High-quality complete target pseudo-labels are generated by using adaptive masking at the feature level, visual prototype alignment, and instance completion correction at the label level. Local overfitting is suppressed by a learnable feature masking module, a semantic alignment mechanism is constructed by introducing a multi-instance learning module for visual prototype alignment, and fragmented pseudo-labels are denoised and spatially fused by an instance completion module.
It significantly improves the localization integrity and classification accuracy of weakly supervised open-vocabulary object detection, enhancing the accuracy and robustness of the object detection model.
Smart Images
Figure CN122116390B_ABST