基于分层混合网络的底层视觉颜色成像学习方法和装置
By using the Color-HHN model with a hierarchical hybrid network, combined with the CLIP large model and multi-task learning, the problem of labeling low-level color tasks is solved, and efficient processing and generalization ability of various color tasks are achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- RENMIN ZHONGKE (JINAN) INTELLIGENT TECH CO LTD
- Filing Date
- 2025-11-04
- Publication Date
- 2026-07-17
AI Technical Summary
Low-level color tasks face challenges such as difficulty in labeling real images, lack of prior knowledge, and lack of contextual information understanding, resulting in long development cycles, weak generalization ability, and difficulty in transfer and expansion of traditional methods.
The Color-HHN model based on hierarchical hybrid networks is adopted. It utilizes CLIP large model and multi-task learning, and trains feature extraction network, hierarchical hybrid network and feature decoding network through mixed training set. Combined with multi-cue information network, it can achieve unified processing of different color tasks.
It improves the model's generalization ability and efficiency, enabling it to handle multiple low-level color tasks under a single model structure, reducing computational overhead and providing high-level auxiliary information.
Smart Images
Figure CN121353105B_ABST
Abstract
Citation Information
Patent Citations
CN119417717A
CN119649170A