单个图像逆向渲染
By using 2D convolutional neural networks and volumetric rendering techniques, 3D scene representations can be generated from a single 2D image, overcoming the limitations of existing technologies that require multiple images and prior information, and achieving real-time 3D scene generation.
CN115797543BActive Publication Date: 2026-07-17NVIDIA CORP
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- NVIDIA CORP
- Filing Date
- 2022-08-09
- Publication Date
- 2026-07-17
AI Technical Summary
Technical Problem
Existing reverse rendering methods require multiple 2D images and ground truth 3D geometry, cannot run in real time, and rely on target-specific prior information, which limits their ability to generate a complete 3D scene representation.
Method used
By using a single 2D image as input, extracting features using a 2D convolutional neural network, and combining volumetric rendering and multilayer perceptron to generate a 3D representation, real-time inverse rendering without specific prior information is achieved.
Benefits of technology
It enables the generation of 3D scene representations using only a single 2D image, improving the performance and generation efficiency of computing hardware and avoiding a lengthy scene optimization process.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN115797543B_ABST
Abstract
本发明公开了单个图像逆向渲染。单个二维(2D)图像可用作输入以获得2D图像的三维(3D)表示。这是通过由编码器从2D图像提取特征，并利用经训练的2D卷积神经网络(CNN)确定2D图像的3D表示来完成的。然后在3D表示上运行体积渲染以组合一个或更多个观察方向内的特征，并且组合的特征作为输入被提供给针对3D表示内的每个点预测和输出颜色(或多维神经特征)和密度值的多层感知器(MLP)。因此，可以仅使用单个2D图像作为输入来执行单个图像逆向渲染，以创建单个2D图像中的场景的对应3D表示。
Need to check novelty before this filing date? Find Prior Art