| %auto-ignore | |
| \begin{abstract} | |
| \vspace{-5.5mm} | |
| Wires and powerlines are common visual distractions that often undermine the aesthetics of photographs. The manual process of precisely segmenting and removing them is extremely tedious and may take up hours, especially on high-resolution photos where wires may span the entire space. In this paper, we present an automatic wire clean-up system that eases the process of wire segmentation and removal/inpainting to within a few seconds. | |
| We observe several unique challenges: wires are thin, lengthy, and sparse. These are rare properties of subjects that common segmentation tasks cannot handle, especially in high-resolution images. | |
| We thus propose a two-stage method that leverages both global and local contexts to accurately segment wires in high-resolution images efficiently, and a tile-based inpainting strategy to remove the wires given our predicted segmentation masks. We also introduce the first wire segmentation benchmark dataset, \benchmark. Finally, we demonstrate quantitatively and qualitatively that our wire clean-up system enables fully automated wire removal with great generalization to various wire appearances. | |
| %We introduce the novel problem of semantic segmentation for wire-like objects in high-resolution images for photo retouching applications. | |
| %We observe several challenging properties of wire appearances that are rare in most segmentation datasets -- thin, lengthy, and sparse, which recent works in high-resolution semantic segmentation cannot effectively solve. We thus propose a two-stage model that leverages global context and local information to predict accurate wire masks, and design an inference pipeline that efficiently handles high-resolution images. | |
| %To encourage future research, we introduce WireSeg-HR, the first benchmark dataset on wire-like object semantic segmentation for photographic applications. Finally, we show our wire segmentation pipeline enables fully automated wire removal for photo retouching. | |
| % , and show that our model outperforms baseline methods. | |
| % In the end, we demonstrate that with our predicted wire masks, wire-like object removal, a common while tedious photo retouching step, can be made fully automated with high quality results. | |
| % We introduce the novel problem of semantic segmentation for wire-like objects in high-resolution images for photo retouching applications. | |
| % We observe several challenging properties of wire appearances that are rare in most segmentation datasets -- thin, extensive, sparse and discontinuous (due to frequent occlusions). We provide comprehensive analyses on the effect of common semantic segmentation approaches over these wire-like objects and propose a two-stage model to overcome these difficulties. | |
| % To encourage research on tackling challenging properties of wires, we introduce WireSeg-HR, the first benchmark dataset on wire-like object semantic segmentation for photographic applications, and show that our model outperforms baseline methods. | |
| % % In the end, we demonstrate that with our predicted wire masks, wire-like object removal, a common while tedious photo retouching step, can be made fully automated with high quality results. | |
| \end{abstract} | |
| % \vspace{-1.343cm} |