Fully Convolutional Neural Network for Stereoscopic 3D View Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing single-image-based view-generation (SIVG) methods, such as Deep3D, require large memory and computational complexity due to pre-trained CNNs with fully-connected layers, limiting spatial resolution and flexibility in generating stereoscopic 3D images.
Innovation Solution
A fully convolutional neural network (FCN) is trained without fully connected layers to process images of varying resolutions, decoupled into luminance and chrominance networks for efficient 3D image rendering, using convolutional layers and a softmax layer for normalization, allowing for flexible and computationally efficient generation of stereoscopic 3D images from a single image.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If pre-trained CNNs with fully-connected layers are used for SIVG, then prediction performance is improved, but memory space and computational complexity increase significantly
Solution Approach 1:
The patent segments the image processing task into separate luminance and chrominance networks that operate independently. This segmentation allows each network to focus on specific image components, reducing overall computational complexity while maintaining prediction performance for depth estimation and view synthesis.
Solution Approach 2:
The patent extracts and removes the fully-connected layers from the traditional CNN architecture, retaining only the convolutional layers. This extraction eliminates the computational bottleneck and memory space requirements associated with dense layers while preserving the feature extraction capabilities needed for accurate prediction.
2Measurement precision
If pre-trained CNNs with fully-connected layers are used for SIVG, then prediction performance is improved, but memory space increases significantly
Solution Approach 1:
The patent extracts and removes the fully-connected layers from the traditional CNN architecture, retaining only the convolutional layers. This extraction eliminates the computational bottleneck and memory space requirements associated with dense layers while preserving the feature extraction capabilities needed for accurate prediction.
3Measurement precision
If dense fully connected layers are used in the network, then prediction performance is improved, but spatial resolution of input and output is limited
Solution Approach 1:
The patent extracts and removes the fully-connected layers from the traditional CNN architecture, retaining only the convolutional layers. This extraction eliminates the computational bottleneck and memory space requirements associated with dense layers while preserving the feature extraction capabilities needed for accurate prediction.
Data Source
AI summary
A method of rendering a stereoscopic 3D image from a single image, including receiving a collection of pairs of 3D images including an input image and an output image that is a 3D pair of the input image, training a neural network composed of convolutional layers without any fully connected layers, with pairs of 3D images from the collection, to receive an input image and generate an output image that is a 3D pair of the input image, wherein the neural network is provided as an application on a computing device, receiving an input image and generating an output image that is a 3D pair of the input image by the neural network.