language_zcx的博客

Face Forgery Survey

IntroductionSuperPoint: Self-supervise Interest Point Detection and DescriptionAbstractMagicPoint利用简单几何图像的关键点易于提取且定义较为明确的特点来构造简单数据集，并基于此训练简单几何图像兴趣点检测模型MagicPoint。Homographic Adaptation作者利用MS-COCO数据集和上述训练的MagicPoint模型，并结合其提出的Homographic Adaptation机

Large-Scale Object Detection in the Wild from Imbalanced Multi-LabelsPaper PDF文章目录IntorductionInnovationMethodExperimentIntorductionInnovationMethodExperiment

CNN-RNN: A Unified Framework for Multi-label Image ClassificationPaper PDF文章目录IntroductInnovationMethodExperimentIntroductInnovationMethodExperiment

Learning Generalized Spoof Cues for Face Anti-spoofingpaper PDFhttps://blog.youkuaiyun.com/fuwenyan/article/details/77769224

Seeing Voices and Hearing Faces: Cross-modal biometric matchingPaper PDF文章目录IntroductionInnovationMethodExperimentIntroductionInnovationMethodExperiment

Self-supervised Representation Learning from Videos for Facial Action Unit Detection(CVPR 2019)Paper PDF文章目录IntroductionInnovationMethodFeature disentanglingTarget reconstructionCycle with AU/pose changedExperimentImplementation details DetailedCompariso

Face-swap方式，将target的脸替换为source的脸，形成deepfake video.这种方式只是改变了target的型，而没有改变target的神，意思就是说deepfake video中人物的神情习惯仍然是target的。作者利用这一特征。通过将人物identity和behavior feature关联起来。作者分别以以video中的behavior feature和face为线...

Orderless Recurrent Models for Multi-label Classification Paper PDF文章目录IntroductionInnovationMethodImage-to-sequence modelTraining recurrent modelsOrderless recurrent modelsExperimentsIntroductionM...

See Better Before Looking Closer: Weakly Supervised Data Augmentation Netowrk for Fine-Grained Visual ClassificationPaper PDF文章目录AbstractPipelineWeakly Supervised Attention LearningSpatial Represent...

文章目录Noun interpretationRCNNRPNFeature Map上采样下采样高斯金字塔、高斯平滑、亚采样bottom-up网络top-down网络Noun interpretationRCNNRPNFeature Pyramid NetworksFeature Map在每个卷积层，数据都是以三维形式存在的。你可以把它看成许多个二维图片叠在一起，其中每一个称为一个fea...