JoeyF's Home


  • Home

  • About

  • Tags

  • Categories

  • Archives

去噪论文阅读-3

Posted on 2021-02-17
Learning Enriched Features for Real Image Restoration and EnhancementMotivation Full-resolution representation is spatially precise but contextually l ...
Read more »

去噪论文阅读-2-Self Supervised

Posted on 2021-02-17
导言:这篇文章两篇论文,Self Supervised去噪相关。 Neighbor2Neighbor: Self-Supervised Denoising from Single Noisy Images时间线: Noise2Noise Probabilistic Noise2Void Noise ...
Read more »

面经

Posted on 2021-02-17
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 ...
Read more »

去噪论文阅读-1

Posted on 2021-02-17
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 ...
Read more »

Video Object Segmentation (VOS)

Posted on 2020-08-20
实验记录:T0819:对作者提供的模型进行test,得到结果,使用crf对其进行后处理。使用std davis-python代码测试,其结果为: 第一次测试baseline模型结果 对作者提供的结果,使用std davis-python代码进行测试,其结果为: ...
Read more »

Gan Loss improve之LSGan WGan WGan-GP

Posted on 2020-08-15
导言:接上篇经典Gan Loss原理以及其实现,本篇深入探讨一下各种GanLoss。 经典GAN loss缺点: 在经典GAN LOSS中Disciminator是一个Binery Classifier, 对于图中所示的两种分布,二分类判别器给出的两者loss是相同的,但是实际上G_1的损失应该要 ...
Read more »

经典GAN loss原理以及其实现

Posted on 2020-08-15
导言:前几天同门问起我GAN loss的实现,我发现自己在一些符号、细节上对GAN loss还是有没有记牢的地方。于是写下这篇blog来加深印象。 GANS loss原理经典GAN loss是最原始的loss: \min_{G} \max_{D} E_{x \in q(x)}[log D(x)] ...
Read more »

Deepfake Detection之FaceForensics++

Posted on 2020-07-13
Dataset:Face2FaceFacial reenactment method. FaceSwapGraphics-based approach to transfer the face region from a source video to a target video. Based o ...
Read more »

Deepfake Detection之Face Xray

Posted on 2020-07-08
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 ...
Read more »

Deepfake Detection之DeeperForensics-1.0

Posted on 2020-07-08
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 ...
Read more »
1…345…7
JoeyF

JoeyF

61 posts
36 tags
GitHub E-Mail
© 2022 JoeyF
Powered by Hexo
|
Theme — NexT.Gemini v5.1.4