论文笔记
文章平均质量分 60
整理以往所读论文笔记
梦无羡
记录视频编码学习历程,防止老年痴呆!!
展开
-
【论文笔记】(VC)A QP-ADAPTIVE MECHANISM FOR CNN-BASED FILTER IN VIDEO CODING
A QP-ADAPTIVE MECHANISM FOR CNN-BASED FILTER IN VIDEO CODING问题在先前大多数工作中,每个QP带都需要一个单独的模型,参数过多。方法因此,文中提出将QP引入卷积中,而且文中所提结构可以直接替换现有工作的任何卷积层。分析In [20], it introduces the QP as another input.Its working principle as flowes:As wo can see, the bias is wei原创 2021-02-23 17:52:32 · 271 阅读 · 0 评论 -
【论文笔记】(VC)Multi-scale Grouped Dense Network for VVC Intra Coding
Multi-scale Grouped Dense Network for VVC Intra Coding问题为了降低伪影方法提出了一个 the multi-scale grouped dense network (MSGDN)包含了多尺度和组密集模块提出 a generative adversarial network (MSGDNGAN) 利用MSGDN作为生成器,以提高主观质量分析VVC中,编码单元的尺寸是不同的,因此造成压缩伪影分布在不同的空间尺寸上。因此采用多尺度为了增强网络的原创 2021-02-23 10:19:23 · 349 阅读 · 0 评论 -
【论文笔记】(Frequency)FcaNet: Frequency Channel Attention Networks
《FcaNet: Frequency Channel Attention Networks》question常见的注意力机制,都是默认将GAP作为一个不可或缺的预处理过程,然而文中提出GAP是频域中特征分解的一个特例,因此提出在频域进行注意力机制的预处理过程。...原创 2021-02-21 16:55:42 · 760 阅读 · 0 评论 -
【论文笔记】(SR)《Image Super-Resolution Using Very Deep Residual Channel Attention Networks》
《Image Super-Resolution Using Very Deep Residual Channel Attention Networks》阅读笔记problemThe lowresolution inputs and features contain abundant low-frequency information, which is treated equally across channels, hence hindering the representational abilit原创 2020-12-23 10:38:22 · 233 阅读 · 0 评论 -
【论文笔记】(HEVC)《Multi-stage Attention Convolutional Neural Networks for HEVC In-Loop Filtering》
《Multi-stage Attention Convolutional Neural Networks for HEVC In-Loop Filtering》阅读笔记Peng-Ren Lai, Jia-Shung Wang Department of Computer Science, National Tsing Hua University, Hsinchu, Taiwan {ex0487, jswangtaiwan}@gmail.com abstractthey proposed a M原创 2020-11-11 16:14:55 · 386 阅读 · 0 评论 -
【论文笔记】(HEVC)《Recursive Residual Convolutional Neural NetworkBased In-Loop Filtering ..》
《Recursive Residual Convolutional Neural NetworkBased In-Loop Filtering for Intra Frames》阅读笔记problems深层网络需要考虑退化和训练难度的问题motivation提出RRCNN,有利于实现深层网络,在增强学习能力的同时而不需考虑退化问题,递归使用一组权值,利用更少的参数实现好的效果。shortcut connections 降低训练难度。亮度色度分别训练网络以适应不同通道的特点。引入CTU原创 2020-11-11 14:16:09 · 293 阅读 · 0 评论 -
【论文笔记】(VC) 《Frame-wise CNN-based Filtering for Intra-Frame Quality Enhancement of HEVC Videos》
《Frame-wise CNN-based Filtering for Intra-Frame Quality Enhancement of HEVC Videos》阅读笔记question基于块的训练和滤波不太好,有一些缺点motivation提出了一种基于帧的CNN滤波器,就是用整帧来训练网络。基于全局帧内容,网络可以检测整帧图片的低质量区域,compute the necessary details for enhancingthe quality of the frames comp原创 2020-11-06 15:28:36 · 416 阅读 · 0 评论 -
【论文笔记】(VC)《Combining Progressive Rethinking and Collaborative Learning: ... for In-Loop filtering》
《Combining Progressive Rethinking and Collaborative Learning: A Deep Framework for In-Loop Filtering 》阅读笔记quedtion1、how to model spatial and temporal redundancies more effectively in the coding scenario(如何建模,有效的利用时空冗余)2、what kinds of side information原创 2020-11-04 15:35:14 · 367 阅读 · 0 评论 -
【论文笔记】(VC in-loop filter)《MFRNet: A New CNN Architecture for Post-Processing and In-loop Filtering 》
《MFRNet: A New CNN Architecture for Post-Processing and In-loop Filtering 》论文阅读笔记提出的问题莫得!motivation提出了一种multilevel feature review residual dense blocks for postprocessing and in-loop filtering. 其实就是单纯的利用了dense residual net结构,同样的多个MFRB( Multi-level Fea原创 2020-10-30 15:26:13 · 734 阅读 · 4 评论 -
【论文笔记】(SR)《Residual Dense Network for Image Super-Resolution》
《Residual Dense Network for Image Super-Resolution》论文阅读笔记针对的问题 :针对目前大多数基于CNN的SR模型不能充分利用原始低分辨率(LR)图像的层次特征(the hierarchical features ),只实现了较低的表现。提出的办法 :提出了一种 residual dense network (RDN)解决这个问题。使用 residual dense block (RDB) :使用密集连接的卷积层来提取局部特征RDB会将前面的RDB状原创 2020-10-28 09:38:38 · 474 阅读 · 0 评论