基于卷积核重参数化再校准的轻量级图像超分辨率方法
By introducing reparameterized kernel recalibrated convolutions and context feature fusion blocks into a lightweight image super-resolution network, the problems of limited expressive power and narrow receptive field of group convolutional models are solved, and efficient image super-resolution reconstruction is achieved.
CN118674622BActive Publication Date: 2026-07-17UNIV OF ELECTRONICS SCI & TECH OF CHINA
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- UNIV OF ELECTRONICS SCI & TECH OF CHINA
- Filing Date
- 2024-06-06
- Publication Date
- 2026-07-17
AI Technical Summary
Technical Problem
Existing group convolutions suffer from limited model expressive power and narrow receptive field in lightweight image super-resolution networks, which limits their practicality in real-world scenarios.
Method used
By introducing reparameterized kernel recalibrated convolution (RecConv) and efficient context feature fusion block (CFFB), combined with progressive multi-scale recalibration block (PMRB), the model's feature representation capability is enhanced and the receptive field is expanded.
Benefits of technology
Without increasing computational costs, the model's expressive power and performance were significantly improved, the number of parameters was reduced, and better image super-resolution results were achieved.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN118674622B_ABST
Abstract
本发明提供一种基于卷积核重参数化再校准的轻量级图像超分辨率方法，提出了一种新颖的内核重校准策略，将一个通道的内核与其他通道的参数相关联，允许内核重关注他们之前没有关注过的模型的其他部分，而不是只关注独立的输入渠道。基于该设计，我们提出了一种新颖的渐进式多尺度重校准块，以通过各种多尺度感受野捕获更多判别性特征。训练结束后，可以将内核重校准带来的参数重参数化，以与原始卷积对齐。最终生成的块保持与原始块相同的推理成本，但提供了改进的性能。引入了一种轻量级的SR网络，称为内核重校准引导网络，旨在实现性能和适用性之间的卓越平衡。本发明已通过综合实验验证了竞争结果。同时使用更少的参数。
Need to check novelty before this filing date? Find Prior Art