Any-scale image down-sampling method based on meta-learning
The invention discloses an arbitrary-scale image down-sampling method based on meta-learning. The method comprises the following steps: constructing an arbitrary-scale image down-sampling model based on meta-learning, wherein the image down-sampling model comprises a jump connection network and a fo...
Saved in:
Main Authors | , , , , , |
---|---|
Format | Patent |
Language | Chinese English |
Published |
25.02.2022
|
Subjects | |
Online Access | Get full text |
Cover
Loading…
Summary: | The invention discloses an arbitrary-scale image down-sampling method based on meta-learning. The method comprises the following steps: constructing an arbitrary-scale image down-sampling model based on meta-learning, wherein the image down-sampling model comprises a jump connection network and a forward neural network, the jump connection network comprises a single-scale down-sampling network and a bicubic interpolation method, and the forward neural network comprises a feature extraction network and a convolution kernel parameter prediction network based on meta-learning; carrying out element addition operation on the output of the jump connection network and the forward neural network to obtain a low-resolution image; and training the constructed image down-sampling model, and performing image down-sampling through the trained image down-sampling model.
本发明公开了一种基于元学习的任意尺度图像下采样方法,包括如下步骤:基于元学习构建任意尺度图像下采样模型;所述图像下采样模型包括跳跃连接网络和前向神经网络;所述跳跃连接网络包括单尺度下采样网络和双三次插值方法;所述前向神经网络包括特征提取网络和基于元学习的卷积核参数预测网络;所述跳跃连接网络和前向神经网络的输出 |
---|---|
Bibliography: | Application Number: CN202111344541 |