模型改进方法及装置
By performing gradient heatmap analysis and weight adjustment on the intermediate layers of the image processing model, the problem of time-consuming and labor-intensive adjustment of deep learning models is solved, achieving efficient model improvement and accuracy enhancement.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BEIJING JINGWEI HIRAIN TECH CO INC
- Filing Date
- 2024-03-25
- Publication Date
- 2026-07-17
AI Technical Summary
In existing technologies, adjusting the network structure of deep learning models requires traversal adjustments, which is time-consuming and laborious, and may lead to attempts in the wrong or insignificant direction, making it difficult to efficiently improve the model's prediction accuracy.
Gradient analysis is performed by obtaining the intermediate layer output of the image processing model to generate a gradient heatmap. The intermediate layer weights corresponding to the target gradient heatmap are reduced until the model with the highest image processing accuracy is determined, avoiding iterative adjustments.
This enables efficient and targeted adjustment of image processing models without traversal adjustments, improving the model's prediction accuracy and reducing adjustment costs.
Smart Images
Figure CN120707979B_ABST