[TOC] 截至2019年12月11日 2019年度阅读论文: 视觉领域论文Fashion 相关论文:1611.09577 Fast Face-swap Using Convolutional Neural Networks.pdf1711.08447 [VITON] An Image-based Virtual Try-on Network.pdf1807.07688 [CP-VTON] Toward Characteristic-Preserving Image-based Virtual Try-On Network.pdf1906.01347 [WUTON] End-to-End Learning of Geometric Deformations of Feature Maps for Virtual Try-On .pdf1906.07251 (ncsu JD Oppo)Pose Guided Fashion Image Synthesis Using Deep Generative Model.pdf[MG-VTON] Towards Multi-pose Guided ...

Data Gen 1、图片元数据 2、图片标注 安装 1234conda create --name=labelme python=3.6conda activate labelmepython -m pip install pyqt5python -m pip install labelme https://github.com/wkentaro/labelme https://github.com/hitzoro/FCN-ColorLabel Model build Model Train Model inference

[TOC] ckpt 2 pb12345678910111213141516def save_pb(dst_pb, sess, output_node_names=None): # nn build # saver = tf.train.Saver(tf.global_variables()) # sess = tf.Session() # saver.restore(sess, ckpt) constant_graph = graph_util.convert_variables_to_constants(sess, sess.graph_def, output_node_names) with tf.gfile.FastGFile(dst_pb, mode='wb') as f: f.write(constant_graph.SerializeToString()) return constant_graphdef main(): # model restore dst_file = 'dest.pb' na ...

[toc] CNN Geometric 中文介绍论文1: CNN GeometricConvolutional neural network architecture for geometric matching 卷积神经网络结构用于几何匹配 I. Rocco, R. Arandjelović and J. Sivic. Convolutional neural network architecture for geometric matching. CVPR 2017 [website][arXiv] 架构：阶段1：仿射变换 estimates an affine transformation 阶段2：薄板样条转换 thin-plate spline (TPS) transformation Started: demo.py demonstrates the results on the ProposalFlow dataset (Proposal Flow Dataset 的示范结果) train.py is the main training script (训练入口) e ...

CV PoseEstimation Method[TOC] Human Pose Estimation 论文/原理 S-单人D-多人 DensePose 实时人体姿态估计 单人姿态估计 DeepPose 最早应用CNN的方法,直接回归关节坐标 S Flowing ConvNets 热力图->关节点 S CPM multi-stage和refine heatmap的思想 S stacked hourglass 卷积层、解卷积层. S 多人姿态估计 DeepCut CVPR2016 《DeepCut: Joint Subset Partition and Labeling for Multi Person Pose Estimation》 S/D OpenPose Realtime Multi-Person 2D Pose Estimation using Part Affinity Fields S/D PAF Realtime Multi-Person 2D Pose Estimation using Part ...

[toc] 人体解析数据集 -human parsing 开源 项目 LIP [open] 2017 SS-NAN Single Person http://hcp.sysu.edu.cn/lip https://github.com/Engineering-Course/LIP_SSL CIHP [open] 2018 http://sysu-hcp.net/lip/overview.php ATR [open] 2015 Chictopia10k Co-CNN human parsing “Human parsing with contextualized convolutional neural network.” ICCV’15, VIP [open] Video Multi-Person Human Parsing MHP [open] 2017 https://lv-mhp.github.io/dataset Person-Part [open] 2014 http://www.stat.ucla.edu/~xia ...

[toc] CP-VTON1807.07688 https://github.com/sergeywong/cp-vton 解决了在真实的虚拟试穿情况下面临的在大空间变形时的服装细节的保留问题。 通过GMM模块整合了全学习的TPS，用来获得更健壮和更强大的对齐图像。 在给定对齐图像的基础上，通过Try-On模块来动态合并渲染结果与变形结果。 CP-VTON网络的性能已经在Han等人收集的数据集上进行了证明。 Motivation：“在保留目标服装细节的情况下将服装转换为适合目标人物的体型” 几何匹配模块GMM，将目标服装转换为适合目标人物体型的形状 Try-On模块将变形后的服装与人物整合并渲染整合后的图像 架构alignment Network 对齐网络 a single pass generative framework 单通道生成框架 GMM模型 几何匹配模块Geometric Matching Moduel extracting network: IN_cloth-agnostic_person [h, w, 22] (64, 128, 256, 512, ...

[TOC] VTON code 1711.08447 [VITON] An Image-based Virtual Try-on Network.pdf VITON：基于映像的虚拟试穿网络CVPR 2018论文“ VITON：基于图像的虚拟试穿网络”的代码和数据集 人像提取本文中使用的人物表示由2D姿态估计器和人工解析器提取： 实时多人姿势估计 自我监督的结构敏感学习 数据集由于版权问题，该数据集不再公开可用。对于已经下载数据集的用户，请注意，使用或分发数据集是非法的！ This dataset is crawled from women’s tops on Zalando. 测试第一阶段在Google云端硬盘上下载经过预训练的模型。将它们放在model/文件夹下。 运行test_stage1.sh以进行推断。结果在中results/stage1/images/。results/stage1/index.html可视化结果。 第二阶段运行matlab脚本shape_context_warp.m以提取TPS转换控制点。 然后test_stage2.sh进行优化并生成最终结果，该结果 ...

[TOC] MG-VTON 《Towards Multi-pose Guided Virtual Try-on Network 》https://arxiv.org/abs/1902.11026 Virtual try-on system 虚拟试穿系统。 —现有的方法，只能从单个固定的人体姿态上换装。往往失去纹理细节，缺少姿态多样性。 three Stages: 1) a desired human parsing map of the target image is synthesized to match both the desired pose and the desired clothes shape; 合成目标图像的所需人工解析映射，以匹配所需的姿态和所需的衣服形状 2) a deep Warping Generative Adversarial Network (Warp-GAN) warps the desired clothes appearance into the synthesized human parsing map and alleviates the ...