人工智能培训

搜索

人工智能论文:视觉对象网络:使用解缠结的3D表示生成图像(Visual Object Networks: Image Generation with Disen

[复制链接]
rrrrrrr 发表于 4 天前 | 显示全部楼层 |阅读模式
rrrrrrr 4 天前 475 0 显示全部楼层
人工智能论文:视觉对象网络:使用解缠结的3D表示生成图像(Visual Object Networks: Image Generation with Disentangled 3D  Representation)深度生成模型的最新进展已经在图像生成方面取得了巨大的突破。然而,虽然现有模型可以合成真实感图像,但他们缺乏对我们潜在3D世界的理解。我们提出了一种新的生成模型,即视觉对象网络(VON),利用解开的3D表示来合成对象的自然图像。受经典图形渲染管道的启发,我们将图像形成过程解释为三个条件独立的因素 - 形状,视点和纹理 - 并呈现端对端的对抗性学习框架,共同模拟3D形状和2D图像。我们的模型首先学会合成与真实形状无法区分的3D形状。然后,它在采样的视点下从其形状渲染对象的2.5Dsketches(即轮廓和深度图)。最后,它学会了为这些2.5D sketchesto添加逼真的纹理,生成自然图像。 VON不仅生成比最先进的2D图像合成方法更现实的图像,还能够进行多种3D操作,例如改变生成图像的视点,编辑形状和纹理,纹理和形状空间中的线性插值,并且跨越不同的对象和视点转移外观。
Recent progress in deep generative models has led to tremendous breakthroughsin image generation.However, while existing models can synthesizephotorealistic images, they lack an understanding of our underlying 3D world.We present a new generative model, Visual Object Networks (VON), synthesizingnatural images of objects with a disentangled 3D representation.Inspired byclassic graphics rendering pipelines, we unravel our image formation processinto three conditionally independent factors---shape, viewpoint, andtexture---and present an end-to-end adversarial learning framework that jointlymodels 3D shapes and 2D images.Our model first learns to synthesize 3D shapesthat are indistinguishable from real shapes.It then renders the object's 2.5Dsketches (i.e., silhouette and depth map) from its shape under a sampledviewpoint.Finally, it learns to add realistic texture to these 2.5D sketchesto generate natural images.The VON not only generates images that are morerealistic than state-of-the-art 2D image synthesis methods, but also enablesmany 3D operations such as changing the viewpoint of a generated image, editingof shape and texture, linear interpolation in texture and shape space,andtransferring appearance across different objects and viewpoints.人工智能论文:视觉对象网络:使用解缠结的3D表示生成图像(Visual Object Networks: Image Generation with Disentangled 3D  Representation) n6m7ceO266l4dnR7.jpg
URL地址:https://arxiv.org/abs/1812.02725     ----pdf下载地址:https://arxiv.org/pdf/1812.02725    ----人工智能论文:视觉对象网络:使用解缠结的3D表示生成图像(Visual Object Networks: Image Generation with Disentangled 3D  Representation)
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

rrrrrrr当前离线
新手上路

查看:475 | 回复:0

快速回复 返回顶部 返回列表