A coarse-to-fine approach for dynamic-to-static image translation
作者：
Highlights：
• Dynamic-to-static image translation is delicately formulated as an image inpainting-like problem, and a novel coarse-to-fine framework is proposed.
• A simple but effective strategy is designed to handle the presences of object shadows.
• A mutual texture-structure attention module is proposed to enhance the recovery of textures and structures in dynamic areas.
• We generate a new test dataset with high diversity to supplement the existing test dataset.
• Extensive experimental results demonstrate the superiority of our proposed approach.
摘要
•Dynamic-to-static image translation is delicately formulated as an image inpainting-like problem, and a novel coarse-to-fine framework is proposed.•A simple but effective strategy is designed to handle the presences of object shadows.•A mutual texture-structure attention module is proposed to enhance the recovery of textures and structures in dynamic areas.•We generate a new test dataset with high diversity to supplement the existing test dataset.•Extensive experimental results demonstrate the superiority of our proposed approach.
论文关键词：Dynamic-to-static image translation,Shadow detection,Attention mechanism,Visual place recognition
论文评审过程：Received 20 February 2021, Revised 28 September 2021, Accepted 15 October 2021, Available online 16 October 2021, Version of Record 25 October 2021.
论文官网地址：https://doi.org/10.1016/j.patcog.2021.108373
