{"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/recovering-realistic-texture-in-image-super","title":"Recovering Realistic Texture in Image Super-resolution by Deep Spatial Feature Transform","arxiv_id":"1804.02815","date":"2018-04-09","proceeding":"CVPR 2018 6","authors":["Xintao Wang","Ke Yu","Chao Dong","Chen Change Loy"],"abstract":"Despite that convolutional neural networks (CNN) have recently demonstrated\nhigh-quality reconstruction for single-image super-resolution (SR), recovering\nnatural and realistic texture remains a challenging problem. In this paper, we\nshow that it is possible to recover textures faithful to semantic classes. In\nparticular, we only need to modulate features of a few intermediate layers in a\nsingle network conditioned on semantic segmentation probability maps. This is\nmade possible through a novel Spatial Feature Transform (SFT) layer that\ngenerates affine transformation parameters for spatial-wise feature modulation.\nSFT layers can be trained end-to-end together with the SR network using the\nsame loss function. During testing, it accepts an input image of arbitrary size\nand generates a high-resolution image with just a single forward pass\nconditioned on the categorical priors. Our final results show that an SR\nnetwork equipped with SFT can generate more realistic and visually pleasing\ntextures in comparison to state-of-the-art SRGAN and EnhanceNet.","url_abs":"http://arxiv.org/abs/1804.02815v1","url_pdf":"http://arxiv.org/pdf/1804.02815v1.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":"recovering-realistic-texture-in-image-super","repo_url":"https://github.com/xinntao/SFTGAN","is_official":1,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}},{"paper_slug":"recovering-realistic-texture-in-image-super","repo_url":"https://github.com/micmic123/qmapcompression","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null},{"paper_slug":"recovering-realistic-texture-in-image-super","repo_url":"https://github.com/sdauzcm/sr-basicsr","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"Apache-2.0"}},{"paper_slug":"recovering-realistic-texture-in-image-super","repo_url":"https://github.com/xinntao/BasicSR","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}}],"tasks":[{"task_slug":"image-super-resolution","task_name":"Image Super-Resolution"},{"task_slug":"semantic-segmentation","task_name":"Semantic Segmentation"},{"task_slug":"super-resolution","task_name":"Super-Resolution"}],"methods":[{"method_slug":"batch-normalization","method_name":"Batch Normalization"},{"method_slug":"convolution","method_name":"Convolution"},{"method_slug":"dense-connections","method_name":"Dense Connections"},{"method_slug":"dropout","method_name":"Dropout"},{"method_slug":"max-pooling","method_name":"Max Pooling"},{"method_slug":"prelu","method_name":"PReLU"},{"method_slug":"pixelshuffle","method_name":"PixelShuffle"},{"method_slug":"relu","method_name":"ReLU"},{"method_slug":"residual-block","method_name":"Residual Block"},{"method_slug":"residual-connection","method_name":"Residual Connection"},{"method_slug":"spade","method_name":"SPADE"},{"method_slug":"srgan","method_name":"SRGAN"},{"method_slug":"srgan-residual-block","method_name":"SRGAN Residual Block"},{"method_slug":"sigmoid-activation","method_name":"Sigmoid Activation"},{"method_slug":"softmax","method_name":"Softmax"},{"method_slug":"spatial-feature-transform","method_name":"Spatial Feature Transform"},{"method_slug":"vgg-loss","method_name":"VGG Loss"}],"datasets_introduced":[{"slug":"ost300","name":"OST300","full_name":"OST300"}],"methods_introduced":[{"slug":"spatial-feature-transform","name":"Spatial Feature Transform","full_name":"Spatial Feature Transform"}],"results":[{"leaderboard":"/sota/image-super-resolution-on-bsd100-4x-upscaling","task":"Image Super-Resolution","dataset":"BSD100 - 4x upscaling","model":"SFT-GAN","rank_in_archive_order":62,"of":71,"metrics":{"PSNR":"25.33","SSIM":"0.651"},"uses_additional_data":false},{"leaderboard":"/sota/image-super-resolution-on-set14-4x-upscaling","task":"Image Super-Resolution","dataset":"Set14 - 4x upscaling","model":"SFT-GAN","rank_in_archive_order":99,"of":104,"metrics":{"PSNR":"26.13","SSIM":"0.694"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1804.02815","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}