- 博客(13)
- 收藏
- 关注
RSS订阅
原创 JPGNet: Joint Predictive Filtering and Generative Network for Image Inpainting
JPGNet: Joint Predictive Filtering and Generative Network for Image Inpainting：ACM MM 2021IntroductionMotivationPredictive Filtering for Image InpaintingGenerative Network for Image InpaintingIntroduction 本文认为图像修复实际上是一个图像恢复问题，而不是一个纯粹的图像生成任务。后者只关心生成的图像是真
2021-08-08 10:27:49
723
原创 CR-Fill: Generative Image Inpainting with Auxiliary Contexutal Reconstruction
CR-Fill: Generative Image Inpainting with Auxiliary Contexutal ReconstructionGenerative inpainting networkContextual reconstructionGenerative inpainting network 由粗到细的体系结构，类似于DeepFillv2中的体系结构，但CA Layer被移除，而应用了CR Loss。粗网络和精网络是卷积编码器-解码器类型的网络。扩张的卷积层用于扩大感受野
2021-07-16 21:17:19
1270
原创 Texture Memory-Augmented Deep Patch-Based Image Inpainting
Texture Memory-Augmented Deep Patch-Based Image InpaintingIntroductionIntroduction 上下文关注模块采用softmax函数融合所有候选纹理特征进行生成。在我们的实验中，我们观察到**纹理空间中的插值( interpolation in texture space)**可能会导致模糊的输出和不自然的伪像，如图所示。...
2021-07-16 17:12:12
335
原创 Texture Transform Attention for Realistic Image Inpainting
Texture Transform Attention for Realistic Image InpaintingIntroductionApproachTexture Transform AttentionIntroduction 提出了一个纹理变换注意(TTA)模块，以有效地将纹理信息传递给结果。基于补丁的网络方法通过Channel方向softmax计算每个补丁的相似性，并对每个权重求和。但是，如果背景中有许多相似的面片，则基于相似的权重对多个面片求和，结果会显得模糊。为了解决这个问题，我们搜索
2021-07-14 22:39:28
413
1
原创 MySQL服务停止
Error:Can’t connect to MySQL server on ‘localhost’ (10061)MySQL后使用mysql命令时报错 Can’t connect to MySQL server on localhost (10061)，或者用net start mysql 时报服务名无效，一般是因为mysql服务没有启动。管理身份运行cmd.exe，右键用管理员身份运行MySQL 8.0:载MySQL8.0后解压，根目录没有data文件夹和my.ini文件，我们也不要创建
2021-07-14 15:14:24
466
原创 Learning Statistical Texture for Semantic Segmentation（CVPR2021）
Learning Statistical Texture for Semantic SegmentationIntroductionRelated WorksLow-level Features Extraction.MethodOverall StructureQuantization and Counting Operator1-d QCO2-d QCOTexture Enhancement ModuleIntroduction 图像纹理不仅仅是局部结构属性，而且还涉及全局统计性质。结构性的通常指
2021-07-14 14:51:13
3098
原创 Context-Aware Image Inpainting with Learned Semantic Priors_IJCAI202
Context-Aware Image Inpainting with Learned Semantic Priors_IJCAI2021MethodImage EncodingSemantic Priors LearningContext-Aware Image InpaintingMethod 整体架构如图命名为SPL，以突出其学习和利用语义优先的能力。它包含三个主要组件:图像编码器、语义先验学习器和图像解码器。总之，整个管道可以分为两个阶段。在编码阶段，我们从剩余的损坏图像中学习图像特征和语义先
2021-06-22 16:28:52
1116
1
原创 Face Image Inpainting Network based on Dense Multi-Scale Fusion of Local Binary Patterns——2021
Face Image Inpainting Network based on Dense Multi-Scale Fusion of Local Binary Patterns——2021IntroductionMethodDense Multi-Scale Fusion LBP Learning NetworkLossFace Image Inpainting NetworklossExpriment定性定量AblationIntroduction this paper propose a fac
2021-06-10 09:28:42
340
原创 Deep Generative Model for Image Inpainting with Local Binary Pattern Learning and Spatial Attention
Deep Generative Model for Image Inpainting with Local Binary Pattern Learning and Spatial Attention_IEEE MultiMedia_2021IntroductionRelated workLBPMethodLBP Learning NetworkImage Inpainting NetworkSpatial Attention LayerExperimentAblationIntroduction 网络
2021-06-07 11:38:02
342
1
原创 Dynamic Selection Network for Image Inpainting（IEEE TIP）
Dynamic Selection Network for Image InpaintingMethodLoss FunctionExperimentQualitative AnalysisQuantitative AnalysisAblationEffect of RCNImportant Weights in RCN ModuleMethodLoss FunctionExperimentQualitative Analysis Inpainting中模型会出现伪影和失真结构。PCon
2021-05-18 16:44:17
1041
3
原创 模型参数设置
Parameter Set1 Pconv:优化器和学习率batch_size = 6parser.add_argument('--b1', type=float, default=0.0)parser.add_argument('--b2', type=float, default=0.9)parser.add_argument('--gen_lr', type=float, default=0.0002)lr = args.gen_lrG_optimizer = optim.Adam(ge
2021-05-16 19:07:34
748
原创 pytorch 断点继训
源码：GatedConv parser.add_argument('--resume', action='store_true') parser.add_argument('--resume_epoch', type=int, default=0) # Save generator model def save_model_generator(net, epoch, opt): model_name = 'deepfillv2_WGAN_G_epoch%d_
2021-04-06 19:08:59
195
空空如也
空空如也
TA创建的收藏夹 TA关注的收藏夹
TA关注的人