.png){kind=icon}
.png){kind=icon}
Return
Depth-based adaptable image layer prediction using bidirectional depth semantic fusion
.png){kind=icon}
PRE
AI
DOI：10.1007/s00371-024-03430-2.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Image editing is a laborious process that demands considerable effort to achieve visually appealing outcomes. To address this problem, we propose a novel method depth-based adaptable image layer prediction network for multi-object semantic segmentation, localization, and estimation of the relative depths for image layer prediction. The individual objects in an image are available to be extracted and edited without impacting the background or other objects. Our approach depth semantic prediction network utilizes a fusion module to bidirectionally optimize depth estimation as well as semantic segmentation. In addition, refinement network is designed to refine coarse results into fine segmentation. Experimental results demonstrate the salient progress our method has achieved, and that it can preserve spatial hierarchical relationships and avoid category constraints as well. To the best of our knowledge, this paper introduces a novel approach to image layer prediction, which is more adaptive than the editing method focusing on foreground and background extraction.
Keywords:
Image editing
Image layer prediction
Fusion module
Depth estimation
Semantic segmentation
Journal
IF:
2.9
Papers:
4.6K
Citations:
6.5K