1104710ClothFlow：一种基于流的服装人物生成模型0Xintong Han Xiaojun Hu Weilin Huang � Matthew R. Scott MalongTechnologies, Shenzhen, China Shenzhen Malong Arti�cialIntelligence Research Center, Shenzhen, China0{xinhan,xiahu,whuang,mscott}@malong.com0摘要0我们提出了ClothFlow，一种基于外观流的生成模型，用于合成姿势引导的人物图像生成和虚拟试穿。通过估计源图像和目标图像之间的服装区域之间的密集流，ClothFlow有效地建模几何变化并自然地传递外观以合成新的图像，如图1所示。我们通过一个三阶段的框架实现这一目标：1）在目标姿势的条件下，我们首先估计一个人物语义布局，为生成过程提供更丰富的指导。2）基于两个特征金字塔网络，级联流估计网络准确地估计相应服装区域之间的外观匹配。得到的密集流对源图像进行弯曲，以灵活地处理变形。3）最后，一个生成网络以弯曲的服装区域作为输入，渲染目标视图。我们在DeepFashion数据集上进行了大量实验，用于姿势引导的人物图像生成，并在VITON数据集上进行了虚拟试穿任务。强大的定性和定量结果验证了我们方法的有效性。01. 引言0姿势引导的人物生成[28]在众多现实世界的应用中非常重要，特别是在时尚行业中，客户或造型师希望将服装从一个人转移到另一个人身上。最近在图像到图像翻译的生成网络方面取得的进展，启发了研究人员通过将源图像和目标姿势作为输入，然后合成目标图像[28, 31,29]来解决这个问题。然而，服装的非刚性特性可能会导致剧烈的变形和严重的遮挡，这些问题无法得到适当处理[18]，从而限制了它们在目标视图中渲染服装细节（例如图案、图形、标志）的性能。为了解决这个问题，有两种不同的方法0*Weilin Huang为通讯作者。0范式被用于考虑几何变形以实现更好的外观转换，即基于变形的方法和基于DensePose的方法。基于变形的方法[14,39, 36,4]估计一个变换，可以是仿射变换或薄板样条（TPS），以使源图像像素或CNN特征图发生变形，以处理由姿势差异引起的不对齐。然而，尽管这两种几何建模技术取得了很大的改进，但它们只有有限的自由度（例如，仿射变换为6，TPS为2 × 5 ×5，如[39]所示），当出现大的几何变化时，会导致不准确和不自然的变换估计。最近，一些方法[30, 12,42]使用DensePose[1]描述符作为输入，而不是传统的2D关键点，用于姿势引导的人物生成。DensePose能够将2D图像的人体像素映射到3D人体表面，从而能够传达人体的3D几何信息。这使得即使在有大的空间变形的情况下，更容易获得源图像和目标图像之间的纹理对应关系。然而，将2D图像纹理变形到预定义的基于表面的坐标系中会引入伪影。例如，在源图像中看不见的位置可能会产生洞，这需要通过复杂的纹理修复算法来解决。同时，由于DensePose的估计非常具有挑战性，合成结果通常会受到DensePose估计器性能的影响。因此，与基于变形的方法相比，DensePose转换的结果可能看起来不太逼真[30]。为了解决现有方法中的问题，我们提出了ClothFlow，这是一个基于流的生成模型，用于准确估计源图像和目标图像之间的服装变形，以更好地合成穿着衣服的人物。具体而言，ClothFlow包括3个阶段，如图2所示：（1）有条件的布局生成器首先根据目标姿势预测目标人体分割布局。这使得形状和外观的生成分离，使ClothFlow能够生成更具空间连贯性的结果。104720图1：ClothFlow的结果。左：姿势引导的人物图像生成。右：虚拟试穿。ClothFlow根据所需的姿势对源服装区域进行变形。变形后的服装准确地考虑了源图像和目标图像之间的几何变化，还解决了遮挡问题（例如，当手臂和头发遮挡衣服时）和部分可观察性问题（例如，第一个示例中的拉伸裤子区域和第二个示例中的裙子）。因此，可以使用变形后的衣服作为输入生成具有详细服装图案的逼真图像。0（2）生成的布局作为输入传递给我们的服装流估计阶段，该阶段从源服装预测外观流[45]（即，指示可以用于合成目标图像的源图像中的像素的2D坐标向量）到目标服装。然后，根据估计的流将源服装区域进行变形以考虑几何变形。预测的外观流提供了对视觉对应关系的准确估计，并帮助将源服装区域无缝地转移到合成的目标图像中。0（3）最后，一个保留服装的渲染阶段使用生成网络[34]合成目标图像，同时尽量保留变形后的源服装区域的细节。0我们的方法可以被视为一种基于变形的方法。然而，与大多数基于变形的方法不同，这些方法利用具有少量自由度的几何变换，ClothFlow估计了一个密集的流场（例如，2×256×256），从而在捕捉空间变形时具有高度的灵活性和准确性。与明确利用3D的DensePose方法不同0通过将目标服装区域近似为源图像的变形来隐式捕捉几何变换。ClothFlow的主要贡献如下：•我们精确预测了一个外观流场，以级联的方式对齐源图像和目标图像的服装区域。在每个级联阶段中，特征变形模块逐步改进了先前阶段的估计，并更好地近似了所需的空间变形。•在DeepFashion[27]数据集上进行评估，与最先进的方法相比，ClothFlow通过更好地保留细节服装纹理合成了更逼真的姿势引导图像。我们进一步展示了ClothFlow在VITON数据集[14]上实现的有希望的结果，用于虚拟试穿任务。02. 相关工作0基于变形的图像匹配和合成。空间变换网络[19]允许CNN预测空间变换，这启发了许多最近的工作，用于语义上将一个对象变形为另一个对象[23，33]或变形一个AAAB6nicbZBNS8NAEIYn9avWr6pHL4tF8CAlEUGPBS8eK9oPaEPZbDft0s0m7E6EEvoTvHhQxKu/yJv/xm2ag7a+sPDwzgw78waJFAZd99spra1vbG6Vtys7u3v7B9XDo7aJU814i8Uy1t2AGi6F4i0UKHk30ZxGgeSdYHI7r3eeuDYiVo84Tbgf0ZESoWAUrfVgBmZQrbl1NxdZBa+AGhRqDqpf/WHM0ogrZJIa0/PcBP2MahRM8lmlnxqeUDahI96zqGjEjZ/lq87ImXWGJIy1fQpJ7v6eyGhkzDQKbGdEcWyWa3Pzv1ovxfDGz4RKUuSKLT4KU0kwJvO7yVBozlBOLVCmhd2VsDHVlKFNp2JD8JZPXoX2Zd2zfH9Va1wUcZThBE7hHDy4hgbcQRNawGAEz/AKb450Xpx352PRWnKKmWP4I+fzB2F+jcc=AAAB6nicbZBNS8NAEIYn9avWr6pHL4tF8CAlEUGPBS8eK9oPaEPZbDft0s0m7E6EEvoTvHhQxKu/yJv/xm2ag7a+sPDwzgw78waJFAZd99spra1vbG6Vtys7u3v7B9XDo7aJU814i8Uy1t2AGi6F4i0UKHk30ZxGgeSdYHI7r3eeuDYiVo84Tbgf0ZESoWAUrfVgBmZQrbl1NxdZBa+AGhRqDqpf/WHM0ogrZJIa0/PcBP2MahRM8lmlnxqeUDahI96zqGjEjZ/lq87ImXWGJIy1fQpJ7v6eyGhkzDQKbGdEcWyWa3Pzv1ovxfDGz4RKUuSKLT4KU0kwJvO7yVBozlBOLVCmhd2VsDHVlKFNp2JD8JZPXoX2Zd2zfH9Va1wUcZThBE7hHDy4hgbcQRNawGAEz/AKb450Xpx352PRWnKKmWP4I+fzB2F+jcc=AAAB6nicbZBNS8NAEIYn9avWr6pHL4tF8CAlEUGPBS8eK9oPaEPZbDft0s0m7E6EEvoTvHhQxKu/yJv/xm2ag7a+sPDwzgw78waJFAZd99spra1vbG6Vtys7u3v7B9XDo7aJU814i8Uy1t2AGi6F4i0UKHk30ZxGgeSdYHI7r3eeuDYiVo84Tbgf0ZESoWAUrfVgBmZQrbl1NxdZBa+AGhRqDqpf/WHM0ogrZJIa0/PcBP2MahRM8lmlnxqeUDahI96zqGjEjZ/lq87ImXWGJIy1fQpJ7v6eyGhkzDQKbGdEcWyWa3Pzv1ovxfDGz4RKUuSKLT4KU0kwJvO7yVBozlBOLVCmhd2VsDHVlKFNp2JD8JZPXoX2Zd2zfH9Va1wUcZThBE7hHDy4hgbcQRNawGAEz/AKb450Xpx352PRWnKKmWP4I+fzB2F+jcc=AAAB6nicbZBNS8NAEIYn9avWr6pHL4tF8CAlEUGPBS8eK9oPaEPZbDft0s0m7E6EEvoTvHhQxKu/yJv/xm2ag7a+sPDwzgw78waJFAZd99spra1vbG6Vtys7u3v7B9XDo7aJU814i8Uy1t2AGi6F4i0UKHk30ZxGgeSdYHI7r3eeuDYiVo84Tbgf0ZESoWAUrfVgBmZQrbl1NxdZBa+AGhRqDqpf/WHM0ogrZJIa0/PcBP2MahRM8lmlnxqeUDahI96zqGjEjZ/lq87ImXWGJIy1fQpJ7v6eyGhkzDQKbGdEcWyWa3Pzv1ovxfDGz4RKUuSKLT4KU0kwJvO7yVBozlBOLVCmhd2VsDHVlKFNp2JD8JZPXoX2Zd2zfH9Va1wUcZThBE7hHDy4hgbcQRNawGAEz/AKb450Xpx352PRWnKKmWP4I+fzB2F+jcc=Target PoseSyn Seg ˆstAAAB73icbZBNS8NAEIYn9avWr6pHL4tF6KkkIuix4MVjBfsBbSib7aZdutnE3YlQQv+EFw+KePXvePPfuE1z0NYXFh7emWFn3iCRwqDrfjuljc2t7Z3ybmVv/+DwqHp80jFxqhlvs1jGuhdQw6VQvI0CJe8lmtMokLwbTG8X9e4T10bE6gFnCfcjOlYiFIyitXqDCUVihjis1tyGm4usg1dADQq1htWvwShmacQVMkmN6Xtugn5GNQom+bwySA1PKJvSMe9bVDTixs/yfefkwjojEsbaPoUkd39PZDQyZhYFtjOiODGrtYX5X62fYnjjZ0IlKXLFlh+FqSQYk8XxZCQ0ZyhnFijTwu5K2IRqytBGVLEheKsnr0PnsuFZvr+qNetFHGU4g3OogwfX0IQ7aEEbGEh4hld4cx6dF+fd+Vi2lpxi5hT+yPn8AcAEj68=AAAB73icbZBNS8NAEIYn9avWr6pHL4tF6KkkIuix4MVjBfsBbSib7aZdutnE3YlQQv+EFw+KePXvePPfuE1z0NYXFh7emWFn3iCRwqDrfjuljc2t7Z3ybmVv/+DwqHp80jFxqhlvs1jGuhdQw6VQvI0CJe8lmtMokLwbTG8X9e4T10bE6gFnCfcjOlYiFIyitXqDCUVihjis1tyGm4usg1dADQq1htWvwShmacQVMkmN6Xtugn5GNQom+bwySA1PKJvSMe9bVDTixs/yfefkwjojEsbaPoUkd39PZDQyZhYFtjOiODGrtYX5X62fYnjjZ0IlKXLFlh+FqSQYk8XxZCQ0ZyhnFijTwu5K2IRqytBGVLEheKsnr0PnsuFZvr+qNetFHGU4g3OogwfX0IQ7aEEbGEh4hld4cx6dF+fd+Vi2lpxi5hT+yPn8AcAEj68=AAAB73icbZBNS8NAEIYn9avWr6pHL4tF6KkkIuix4MVjBfsBbSib7aZdutnE3YlQQv+EFw+KePXvePPfuE1z0NYXFh7emWFn3iCRwqDrfjuljc2t7Z3ybmVv/+DwqHp80jFxqhlvs1jGuhdQw6VQvI0CJe8lmtMokLwbTG8X9e4T10bE6gFnCfcjOlYiFIyitXqDCUVihjis1tyGm4usg1dADQq1htWvwShmacQVMkmN6Xtugn5GNQom+bwySA1PKJvSMe9bVDTixs/yfefkwjojEsbaPoUkd39PZDQyZhYFtjOiODGrtYX5X62fYnjjZ0IlKXLFlh+FqSQYk8XxZCQ0ZyhnFijTwu5K2IRqytBGVLEheKsnr0PnsuFZvr+qNetFHGU4g3OogwfX0IQ7aEEbGEh4hld4cx6dF+fd+Vi2lpxi5hT+yPn8AcAEj68=AAAB73icbZBNS8NAEIYn9avWr6pHL4tF6KkkIuix4MVjBfsBbSib7aZdutnE3YlQQv+EFw+KePXvePPfuE1z0NYXFh7emWFn3iCRwqDrfjuljc2t7Z3ybmVv/+DwqHp80jFxqhlvs1jGuhdQw6VQvI0CJe8lmtMokLwbTG8X9e4T10bE6gFnCfcjOlYiFIyitXqDCUVihjis1tyGm4usg1dADQq1htWvwShmacQVMkmN6Xtugn5GNQom+bwySA1PKJvSMe9bVDTixs/yfefkwjojEsbaPoUkd39PZDQyZhYFtjOiODGrtYX5X62fYnjjZ0IlKXLFlh+FqSQYk8XxZCQ0ZyhnFijTwu5K2IRqytBGVLEheKsnr0PnsuFZvr+qNetFHGU4g3OogwfX0IQ7aEEbGEh4hld4cx6dF+fd+Vi2lpxi5hT+yPn8AcAEj68=AAAB6nicbZBNS8NAEIYn9avWr6pHL4tF6KkkItRjwYveKtoPaEPZbDft0s0m7E6EEvoTvHhQxKu/yJv/xm2bg7a+sPDwzgw78waJFAZd99spbGxube8Ud0t7+weHR+Xjk7aJU814i8Uy1t2AGi6F4i0UKHk30ZxGgeSdYHIzr3eeuDYiVo84Tbgf0ZESoWAUrfVwNzCDcsWtuQuRdfByqECu5qD81R/GLI24QiapMT3PTdDPqEbBJJ+V+qnhCWUTOuI9i4pG3PjZYtUZubDOkISxtk8hWbi/JzIaGTONAtsZURyb1drc/K/WSzG89jOhkhS5YsuPwlQSjMn8bjIUmjOUUwuUaWF3JWxMNWVo0ynZELzVk9ehfVnzLN9fVRrVPI4inME5VMGDOjTgFprQAgYjeIZXeHOk8+K8Ox/L1oKTz5zCHzmfPyBOjZk=AAAB6nicbZBNS8NAEIYn9avWr6pHL4tF6KkkItRjwYveKtoPaEPZbDft0s0m7E6EEvoTvHhQxKu/yJv/xm2bg7a+sPDwzgw78waJFAZd99spbGxube8Ud0t7+weHR+Xjk7aJU814i8Uy1t2AGi6F4i0UKHk30ZxGgeSdYHIzr3eeuDYiVo84Tbgf0ZESoWAUrfVwNzCDcsWtuQuRdfByqECu5qD81R/GLI24QiapMT3PTdDPqEbBJJ+V+qnhCWUTOuI9i4pG3PjZYtUZubDOkISxtk8hWbi/JzIaGTONAtsZURyb1drc/K/WSzG89jOhkhS5YsuPwlQSjMn8bjIUmjOUUwuUaWF3JWxMNWVo0ynZELzVk9ehfVnzLN9fVRrVPI4inME5VMGDOjTgFprQAgYjeIZXeHOk8+K8Ox/L1oKTz5zCHzmfPyBOjZk=AAAB6nicbZBNS8NAEIYn9avWr6pHL4tF6KkkItRjwYveKtoPaEPZbDft0s0m7E6EEvoTvHhQxKu/yJv/xm2bg7a+sPDwzgw78waJFAZd99spbGxube8Ud0t7+weHR+Xjk7aJU814i8Uy1t2AGi6F4i0UKHk30ZxGgeSdYHIzr3eeuDYiVo84Tbgf0ZESoWAUrfVwNzCDcsWtuQuRdfByqECu5qD81R/GLI24QiapMT3PTdDPqEbBJJ+V+qnhCWUTOuI9i4pG3PjZYtUZubDOkISxtk8hWbi/JzIaGTONAtsZURyb1drc/K/WSzG89jOhkhS5YsuPwlQSjMn8bjIUmjOUUwuUaWF3JWxMNWVo0ynZELzVk9ehfVnzLN9fVRrVPI4inME5VMGDOjTgFprQAgYjeIZXeHOk8+K8Ox/L1oKTz5zCHzmfPyBOjZk=AAAB2XicbZDNSgMxFIXv1L86Vq1rN8EiuCozbnQpuHFZwbZCO5RM5k4bmskMyR2hDH0BF25EfC93vo3pz0JbDwQ+zknIvSculLQUBN9ebWd3b/+gfugfNfzjk9Nmo2fz0gjsilzl5jnmFpXU2CVJCp8LgzyLFfbj6f0i77+gsTLXTzQrMMr4WMtUCk7O6oyaraAdLMW2IVxDC9YaNb+GSS7KDDUJxa0dhEFBUcUNSaFw7g9LiwUXUz7GgUPNM7RRtRxzzi6dk7A0N+5oYkv394uKZ9bOstjdzDhN7Ga2MP/LBiWlt1EldVESarH6KC0Vo5wtdmaJNChIzRxwYaSblYkJN1yQa8Z3HYSbG29D77odOn4MoA7ncAFXEMIN3MEDdKALAhJ4hXdv4r15H6uuat66tDP4I+/zBzjGijg=AAAB33icbZDNSgMxFIXv1L9aq1a3boJFcFVm3OhScKO7irYW2qFk0jttaCYzJHeEMvQR3LhQxLdy59uY/iy09UDg45yE3HuiTElLvv/tlTY2t7Z3yruVver+wWHtqNq2aW4EtkSqUtOJuEUlNbZIksJOZpAnkcKnaHwzy5+e0ViZ6keaZBgmfKhlLAUnZz3c9W2/Vvcb/lxsHYIl1GGpZr/21RukIk9Qk1Dc2m7gZxQW3JAUCqeVXm4x42LMh9h1qHmCNizmo07ZmXMGLE6NO5rY3P39ouCJtZMkcjcTTiO7ms3M/7JuTvFVWEid5YRaLD6Kc8UoZbO92UAaFKQmDrgw0s3KxIgbLsi1U3ElBKsrr0P7ohE4vvehDCdwCucQwCVcwy00oQUChvACb/DuKe/V+1jUVfKWvR3DH3mfPwv5jF4=AAAB33icbZDNSgMxFIXv1L9aq1a3boJFcFVm3OhScKO7irYW2qFk0jttaCYzJHeEMvQR3LhQxLdy59uY/iy09UDg45yE3HuiTElLvv/tlTY2t7Z3yruVver+wWHtqNq2aW4EtkSqUtOJuEUlNbZIksJOZpAnkcKnaHwzy5+e0ViZ6keaZBgmfKhlLAUnZz3c9W2/Vvcb/lxsHYIl1GGpZr/21RukIk9Qk1Dc2m7gZxQW3JAUCqeVXm4x42LMh9h1qHmCNizmo07ZmXMGLE6NO5rY3P39ouCJtZMkcjcTTiO7ms3M/7JuTvFVWEid5YRaLD6Kc8UoZbO92UAaFKQmDrgw0s3KxIgbLsi1U3ElBKsrr0P7ohE4vvehDCdwCucQwCVcwy00oQUChvACb/DuKe/V+1jUVfKWvR3DH3mfPwv5jF4=AAAB6nicbZBNS8NAEIYn9avWr6hHL4tF6KkkXvRY8KK3ivYD2lA22027dLMJuxOhhP4ELx4U8eov8ua/cdvmoK0vLDy8M8POvGEqhUHP+3ZKG5tb2zvl3cre/sHhkXt80jZJphlvsUQmuhtSw6VQvIUCJe+mmtM4lLwTTm7m9c4T10Yk6hGnKQ9iOlIiEoyitR7uBmbgVr26txBZB7+AKhRqDtyv/jBhWcwVMkmN6fleikFONQom+azSzwxPKZvQEe9ZVDTmJsgXq87IhXWGJEq0fQrJwv09kdPYmGkc2s6Y4tis1ubmf7VehtF1kAuVZsgVW34UZZJgQuZ3k6HQnKGcWqBMC7srYWOqKUObTsWG4K+evA7ty7pv+d6rNmpFHGU4g3OogQ9X0IBbaEILGIzgGV7hzZHOi/PufCxbS04xcwp/5Hz+AB8OjZU=AAAB6nicbZBNS8NAEIYn9avWr6pHL4tF6KkkItRjwYveKtoPaEPZbDft0s0m7E6EEvoTvHhQxKu/yJv/xm2bg7a+sPDwzgw78waJFAZd99spbGxube8Ud0t7+weHR+Xjk7aJU814i8Uy1t2AGi6F4i0UKHk30ZxGgeSdYHIzr3eeuDYiVo84Tbgf0ZESoWAUrfVwNzCDcsWtuQuRdfByqECu5qD81R/GLI24QiapMT3PTdDPqEbBJJ+V+qnhCWUTOuI9i4pG3PjZYtUZubDOkISxtk8hWbi/JzIaGTONAtsZURyb1drc/K/WSzG89jOhkhS5YsuPwlQSjMn8bjIUmjOUUwuUaWF3JWxMNWVo0ynZELzVk9ehfVnzLN9fVRrVPI4inME5VMGDOjTgFprQAgYjeIZXeHOk8+K8Ox/L1oKTz5zCHzmfPyBOjZk=AAAB6nicbZBNS8NAEIYn9avWr6pHL4tF6KkkItRjwYveKtoPaEPZbDft0s0m7E6EEvoTvHhQxKu/yJv/xm2bg7a+sPDwzgw78waJFAZd99spbGxube8Ud0t7+weHR+Xjk7aJU814i8Uy1t2AGi6F4i0UKHk30ZxGgeSdYHIzr3eeuDYiVo84Tbgf0ZESoWAUrfVwNzCDcsWtuQuRdfByqECu5qD81R/GLI24QiapMT3PTdDPqEbBJJ+V+qnhCWUTOuI9i4pG3PjZYtUZubDOkISxtk8hWbi/JzIaGTONAtsZURyb1drc/K/WSzG89jOhkhS5YsuPwlQSjMn8bjIUmjOUUwuUaWF3JWxMNWVo0ynZELzVk9ehfVnzLN9fVRrVPI4inME5VMGDOjTgFprQAgYjeIZXeHOk8+K8Ox/L1oKTz5zCHzmfPyBOjZk=AAAB6nicbZBNS8NAEIYn9avWr6pHL4tF6KkkItRjwYveKtoPaEPZbDft0s0m7E6EEvoTvHhQxKu/yJv/xm2bg7a+sPDwzgw78waJFAZd99spbGxube8Ud0t7+weHR+Xjk7aJU814i8Uy1t2AGi6F4i0UKHk30ZxGgeSdYHIzr3eeuDYiVo84Tbgf0ZESoWAUrfVwNzCDcsWtuQuRdfByqECu5qD81R/GLI24QiapMT3PTdDPqEbBJJ+V+qnhCWUTOuI9i4pG3PjZYtUZubDOkISxtk8hWbi/JzIaGTONAtsZURyb1drc/K/WSzG89jOhkhS5YsuPwlQSjMn8bjIUmjOUUwuUaWF3JWxMNWVo0ynZELzVk9ehfVnzLN9fVRrVPI4inME5VMGDOjTgFprQAgYjeIZXeHOk8+K8Ox/L1oKTz5zCHzmfPyBOjZk=AAAB6nicbZBNS8NAEIYn9avWr6pHL4tF6KkkItRjwYveKtoPaEPZbDft0s0m7E6EEvoTvHhQxKu/yJv/xm2bg7a+sPDwzgw78waJFAZd99spbGxube8Ud0t7+weHR+Xjk7aJU814i8Uy1t2AGi6F4i0UKHk30ZxGgeSdYHIzr3eeuDYiVo84Tbgf0ZESoWAUrfVwNzCDcsWtuQuRdfByqECu5qD81R/GLI24QiapMT3PTdDPqEbBJJ+V+qnhCWUTOuI9i4pG3PjZYtUZubDOkISxtk8hWbi/JzIaGTONAtsZURyb1drc/K/WSzG89jOhkhS5YsuPwlQSjMn8bjIUmjOUUwuUaWF3JWxMNWVo0ynZELzVk9ehfVnzLN9fVRrVPI4inME5VMGDOjTgFprQAgYjeIZXeHOk8+K8Ox/L1oKTz5zCHzmfPyBOjZk=AAAB6nicbZBNS8NAEIYn9avWr6pHL4tF6KkkItRjwYveKtoPaEPZbDft0s0m7E6EEvoTvHhQxKu/yJv/xm2bg7a+sPDwzgw78waJFAZd99spbGxube8Ud0t7+weHR+Xjk7aJU814i8Uy1t2AGi6F4i0UKHk30ZxGgeSdYHIzr3eeuDYiVo84Tbgf0ZESoWAUrfVwNzCDcsWtuQuRdfByqECu5qD81R/GLI24QiapMT3PTdDPqEbBJJ+V+qnhCWUTOuI9i4pG3PjZYtUZubDOkISxtk8hWbi/JzIaGTONAtsZURyb1drc/K/WSzG89jOhkhS5YsuPwlQSjMn8bjIUmjOUUwuUaWF3JWxMNWVo0ynZELzVk9ehfVnzLN9fVRrVPI4inME5VMGDOjTgFprQAgYjeIZXeHOk8+K8Ox/L1oKTz5zCHzmfPyBOjZk=AAAB6nicbZBNS8NAEIYn9avWr6pHL4tF6KkkItRjwYveKtoPaEPZbDft0s0m7E6EEvoTvHhQxKu/yJv/xm2bg7a+sPDwzgw78waJFAZd99spbGxube8Ud0t7+weHR+Xjk7aJU814i8Uy1t2AGi6F4i0UKHk30ZxGgeSdYHIzr3eeuDYiVo84Tbgf0ZESoWAUrfVwNzCDcsWtuQuRdfByqECu5qD81R/GLI24QiapMT3PTdDPqEbBJJ+V+qnhCWUTOuI9i4pG3PjZYtUZubDOkISxtk8hWbi/JzIaGTONAtsZURyb1drc/K/WSzG89jOhkhS5YsuPwlQSjMn8bjIUmjOUUwuUaWF3JWxMNWVo0ynZELzVk9ehfVnzLN9fVRrVPI4inME5VMGDOjTgFprQAgYjeIZXeHOk8+K8Ox/L1oKTz5zCHzmfPyBOjZk=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
我的内容管理
展开
