一种基于物理约束和语言提示的工业缺陷生成方法
By designing an open scene feature optimization module, a multi-cue feature optimization module, and an attention-driven physical modeling module, the problem that existing generation methods cannot generate industrial defects that conform to textual and physical constraints in open scenes is solved, and high-accuracy defect generation is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHANDONG UNIV
- Filing Date
- 2025-12-11
- Publication Date
- 2026-07-17
AI Technical Summary
Existing technologies struggle to generate physically constrained industrial defects that are consistent with text descriptions in open scenarios, especially in effectively simulating new objects and defects. Furthermore, existing methods ignore the material's response to illumination and fine-grained semantics, resulting in significant differences between the synthesized results and real images.
The design includes an open scene feature optimization module, a multi-cue feature optimization module, and an attention-driven physical modeling module. Through techniques such as open vocabulary contrast loss, pixel scorer, dynamic mask generator, denoiser, image defect embedding optimization function, and sparse attention map, an industrial defect generation model is constructed to generate defects that conform to text descriptions and physical constraints.
It improves the accuracy of industrial defect generation, enabling the generation of defect images that conform to textual descriptions and physical constraints without relying on defect samples, and adapts to new objects and defects in open scenes.
Smart Images
Figure CN121962793B_ABST