Open vocabulary semantic segmentation method and system based on instance semantic prototype guidance
By using an instance semantic prototype-guided approach, instance masks are generated and visual lexical associations are constrained. Instance semantic prototypes are then constructed for correction and propagation refinement, solving the problems of instance-level semantic instability and boundary inconsistency in open-vocabulary semantic segmentation, and improving the stability and accuracy of the segmentation results.
Patent Information
- Application Number
- CN202610514985.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-04-17
- Publication Date
- 2026-07-17
AI Technical Summary
Existing open-vocabulary semantic segmentation methods suffer from problems such as instance-level semantic instability, strong cross-instance interference, and inconsistent local view reasoning in complex scenarios, making it difficult to effectively solve the problems of semantic breakage and boundary fragmentation within instances.
By using an instance-based semantic prototype-guided approach, an instance mask is generated. The instance mask and auxiliary visual features are used to constrain the scope and strength of visual word associations. An instance semantic prototype is then constructed for guided correction and progressive propagation refinement, achieving cross-view consistency fusion.
It improves regional consistency and boundary quality, enhances the semantic discrimination ability of open lexical terms, reduces the risk of error propagation, and improves the stability and accuracy of segmentation results.
Smart Images
Figure CN122416016A_ABST