{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/ensnet-ensconce-text-in-the-wild","title":"EnsNet: Ensconce Text in the Wild","arxiv_id":"1812.00723","date":"2018-12-03","proceeding":null,"authors":["Shuaitao Zhang","Yuliang Liu","Lianwen Jin","Yaoxiong Huang","Songxuan Lai"],"abstract":"A new method is proposed for removing text from natural images. The challenge\nis to first accurately localize text on the stroke-level and then replace it\nwith a visually plausible background. Unlike previous methods that require\nimage patches to erase scene text, our method, namely ensconce network\n(EnsNet), can operate end-to-end on a single image without any prior knowledge.\nThe overall structure is an end-to-end trainable FCN-ResNet-18 network with a\nconditional generative adversarial network (cGAN). The feature of the former is\nfirst enhanced by a novel lateral connection structure and then refined by four\ncarefully designed losses: multiscale regression loss and content loss, which\ncapture the global discrepancy of different level features; texture loss and\ntotal variation loss, which primarily target filling the text region and\npreserving the reality of the background. The latter is a novel local-sensitive\nGAN, which attentively assesses the local consistency of the text erased\nregions. Both qualitative and quantitative sensitivity experiments on synthetic\nimages and the ICDAR 2013 dataset demonstrate that each component of the EnsNet\nis essential to achieve a good performance. Moreover, our EnsNet can\nsignificantly outperform previous state-of-the-art methods in terms of all\nmetrics. In addition, a qualitative experiment conducted on the SMBNet dataset\nfurther demonstrates that the proposed method can also preform well on general\nobject (such as pedestrians) removal tasks. EnsNet is extremely fast, which can\npreform at 333 fps on an i5-8600 CPU device.","url_abs":"http://arxiv.org/abs/1812.00723v1","url_pdf":"http://arxiv.org/pdf/1812.00723v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"ensnet-ensconce-text-in-the-wild","repo_url":"https://github.com/HCIILAB/Scene-Text-Removal","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"mxnet","reach":{"status":"ok"}},{"paper_slug":"ensnet-ensconce-text-in-the-wild","repo_url":"https://github.com/neouyghur/One-stage-Mask-based-Scene-Text-Eraser","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}},{"paper_slug":"ensnet-ensconce-text-in-the-wild","repo_url":"https://github.com/youdao-ai/SRNet","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok","spdx":"GPL-3.0"}}],"tasks":[{"task_slug":null,"task_name":"CPU"},{"task_slug":null,"task_name":"Generative Adversarial Network"},{"task_slug":"image-text-removal","task_name":"Image Text Removal"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":"https://syntology.ai/paper/1812.00723","atlas_url":"https://app.syntology.ai/?focus=1812.00723","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}