添加要改颜色的字体
添加要改颜色的字体
添加要改颜色的字体
添加要改颜色的字体
本文发表于CVPR20 上
文献地址：
GitHub：
一、疑问：
1、unsupervised 即self-supervised ?
二、解决问题：
ZSSR has a few limitations
-
it requires thousands of backpropagation gradient updates at test time, which requires considerable time to get the result.
-
cannot fully exploit the large-scale external dataset, and rather it depends only on internal structure and patterns, which lacks in the number of total examples.
三、怎么做：
-
meta-training step
make the model adapt fast to new blur
kernel scenarios. -
transfer learning
fully utilize external samples
meta-learning
plays a role in learning task-level knowledge with different downsampling kernels as different tasks.
meta-test step
simple self-supervised learning is conducted to learn image specific
information within a few gradient steps
网络结构
-
large-scale training,
在DIV2K 数据集上利用bicubic 下采样的 LR 图像和 HR 图像对训练
目的是
网络更容易训练得到
缓解MAML训练不稳定问题 -
meta-transfer learning
目标是
we seek to find a sensitive and transferable initial point of the parameter space where a few gradient updates lead to large performance improvements.
meta-training : external dataset
meta-test: internal learning
- meta-test
四、原理：
总结：
一、创新点：
二、借鉴之处：
三、不足之处