一种面向电商直播虚拟试衣的实时语义分割方法
By employing a multi-resolution feature collaborative optimization strategy and a dynamic feature fusion mechanism, the problem of insufficient multi-scale contextual information capture in e-commerce live streaming virtual try-on was solved, achieving efficient real-time semantic segmentation and virtual try-on functions, and improving segmentation accuracy and robustness.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HUNAN CHEM VOCATIONAL TECH COLLEGE
- Filing Date
- 2025-11-17
- Publication Date
- 2026-07-17
AI Technical Summary
In existing e-commerce live streaming virtual try-on scenarios, real-time semantic segmentation methods struggle to fully capture multi-scale contextual information, resulting in insufficient segmentation accuracy and robustness, and model performance is limited by expert experience.
A multi-resolution feature collaborative optimization strategy is adopted. Features are extracted layer by layer through the backbone network. Combined with differentiable feature aggregation gating unit module and dynamic feature fusion mechanism, the fusion ratio of features at different resolutions is adaptively adjusted. The feature aggregation is optimized by using a hierarchical shared neural architecture search space, thereby improving the feature fusion capability.
It enables efficient real-time segmentation of model and clothing areas in e-commerce live streaming scenarios, improving segmentation accuracy and robustness, and providing an efficient virtual try-on experience.
Smart Images
Figure CN121505264B_ABST