人工智能培训

搜索

深度学习论文:LOGAN:潜在超完备空间中的不成对形状变换(LOGAN: Unpaired Shape Transform in Latent Overcomp

[复制链接]
kjdshfjsfgsfg 发表于 2019-3-26 11:49:06 | 显示全部楼层 |阅读模式
kjdshfjsfgsfg 2019-3-26 11:49:06 378 0 显示全部楼层
深度学习论文:LOGAN:潜在超完备空间中的不成对形状变换(LOGAN: Unpaired Shape Transform in Latent Overcomplete Space)我们提出了LOGAN,一个深度神经网络,旨在从不成对的领域学习通用的形状转换。网络是在两组形状上训练的,例如桌子和椅子,但是在两个域中的形状之间既没有配对来监督形状转换,也没有在任何形状之间的任何点对应关系。经过训练,LOGAN从onedomain获取一个形状并将其转换为另一个。我们的网络由一个自动编码组成,将来自两个输入域的形状编码为一个共同的潜在空间,其中潜在的代码以过度完整的方式编码多尺度形状特征。翻译器基于生成对抗网络(GAN),在潜在空间中运行,其中对抗性损失强制执行跨域转换,而特征保留损失确保为自然形状转换保留正确的形状特征。我们进行各种消融研究,以验证我们的每个关键网络设计,并展示基于基线和最新方法的各种示例的非配对形状变换的卓越能力。我们展示了我们的网络能够学习在形状变换期间保留的whatshape功能,无论是本地还是非本地,无论是内容还是样式等,仅取决于输入域对。
We present LOGAN, a deep neural network aimed at learning generic shapetransforms from unpaired domains.The network is trained on two sets of shapes,e.g., tables and chairs, but there is neither a pairing between shapes in thetwo domains to supervise the shape translation nor any point-wisecorrespondence between any shapes.Once trained, LOGAN takes a shape from onedomain and transforms it into the other.Our network consists of an autoencoderto encode shapes from the two input domains into a common latent space, wherethe latent codes encode multi-scale shape features in an overcomplete manner.The translator is based on a generative adversarial network (GAN), operating inthe latent space, where an adversarial loss enforces cross-domain translationwhile a feature preservation loss ensures that the right shape features arepreserved for a natural shape transform.We conduct various ablation studies tovalidate each of our key network designs and demonstrate superior capabilitiesin unpaired shape transforms on a variety of examples over baselines andstate-of-the-art approaches.We show that our network is able to learn whatshape features to preserve during shape transforms, either local or non-local,whether content or style, etc., depending solely on the input domain pairs.深度学习论文:LOGAN:潜在超完备空间中的不成对形状变换(LOGAN: Unpaired Shape Transform in Latent Overcomplete Space) a7q9N0A0RG76tgxT.jpg
URL地址:https://arxiv.org/abs/1903.10170     ----pdf下载地址:https://arxiv.org/pdf/1903.10170    ----深度学习论文:LOGAN:潜在超完备空间中的不成对形状变换(LOGAN: Unpaired Shape Transform in Latent Overcomplete Space)
回复

使用道具 举报

您需要登录后才可以回帖 登录 | 立即注册

本版积分规则 返回列表 发新帖

kjdshfjsfgsfg当前离线
新手上路

查看:378 | 回复:0

快速回复 返回顶部 返回列表