Learning Enriched Features for Real Image Restoration and EnhancementMotivation
Full-resolution representation is spatially precise but contextually l
...
剑指犀牛鸟问答题Attention
Attention机制 SEnet (注意力机制了解吗、SENet)由输入 X 得到输出feature U 以后,对U的每个通道进行全局平均池化得到通道描述子(Squeeze),再利用两层 FC 得到每个通道的权重值,对 U 按通道进行重新加权得到最终输出(Exc
...
导言:这篇文章两篇论文,关于blind去噪相关。
When AWGN-based Denoiser Meets Real NoisesWhat is the problem:
实际噪声是空域相关的,但是生成的AWGN是空域无关的。使用在AWGN中训练的网络在无法在实际noise中获得较好得结果。
T
...
Dataset:Face2FaceFacial reenactment method.
FaceSwapGraphics-based approach to transfer the face region from a source video to a target video. Based o
...
Method and Motivation:通过fake face 与真实face background混合的blending 阶段来进行检验。模糊嵌入存在边界Inconsistent
BI Dataset:Training data is generated from real images. D
...
数据集组成数据集含有1000个manipulate videos以及1000个源视频(FF++的youtube视频)
1000个manipulate 被5个等级的random-type distortions 处理,得到5个1000视频1000个manipulate 使用more than one
...