Hybrid dual stream blender for wide baseline view synthesis
作者：
Highlights：
• Free navigation systems allow users to freely browse a scene by arbitrarily changing viewpoint.
• Missing viewpoints are usually synthesized exploiting the captured neighboring view.
• Free navigation of a scene requires warping some reference views to some desired target viewpoint and blending them to synthesize a virtual view.
• Preliminary view warping allows reducing the size of the convolutional kernels and thus the learnable parameters count.
• A residual encoder–decoder can be used for image blending with a Siamese encoder to further keep the parameters count low.
摘要
•Free navigation systems allow users to freely browse a scene by arbitrarily changing viewpoint.•Missing viewpoints are usually synthesized exploiting the captured neighboring view.•Free navigation of a scene requires warping some reference views to some desired target viewpoint and blending them to synthesize a virtual view.•Preliminary view warping allows reducing the size of the convolutional kernels and thus the learnable parameters count.•A residual encoder–decoder can be used for image blending with a Siamese encoder to further keep the parameters count low.
论文关键词：View synthesis,View blending,Convolutional Neural Network,Hole inpainting
论文评审过程：Received 4 December 2020, Revised 21 June 2021, Accepted 23 June 2021, Available online 30 June 2021, Version of Record 3 July 2021.
论文官网地址：https://doi.org/10.1016/j.image.2021.116366
