{"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/swinia-self-supervised-blind-spot-image","title":"SwinIA: Self-Supervised Blind-Spot Image Denoising without Convolutions","arxiv_id":"2305.05651","date":"2023-05-09","proceeding":null,"authors":["Mikhail Papkov","Pavel Chizhov","Leopold Parts"],"abstract":"Self-supervised image denoising implies restoring the signal from a noisy image without access to the ground truth. State-of-the-art solutions for this task rely on predicting masked pixels with a fully-convolutional neural network. This most often requires multiple forward passes, information about the noise model, or intricate regularization functions. In this paper, we propose a Swin Transformer-based Image Autoencoder (SwinIA), the first fully-transformer architecture for self-supervised denoising. The flexibility of the attention mechanism helps to fulfill the blind-spot property that convolutional counterparts normally approximate. SwinIA can be trained end-to-end with a simple mean squared error loss without masking and does not require any prior knowledge about clean data or noise distribution. Simple to use, SwinIA establishes the state of the art on several common benchmarks.","url_abs":"https://arxiv.org/abs/2305.05651v2","url_pdf":"https://arxiv.org/pdf/2305.05651v2.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":[],"tasks":[{"task_slug":"color-image-denoising","task_name":"Color Image Denoising"},{"task_slug":"denoising","task_name":"Denoising"},{"task_slug":"grayscale-image-denoising","task_name":"Grayscale Image Denoising"},{"task_slug":"image-denoising","task_name":"Image Denoising"},{"task_slug":"medical-image-denoising","task_name":"Medical Image Denoising"}],"methods":[{"method_slug":"attention","method_name":"Attention"},{"method_slug":"dense-connections","method_name":"Dense Connections"},{"method_slug":"layer-normalization","method_name":"Layer Normalization"},{"method_slug":"linear-layer","method_name":"Linear Layer"},{"method_slug":"multi-head-attention","method_name":"Multi-Head Attention"},{"method_slug":"relative-position-encodings","method_name":"Relative Position Encodings"},{"method_slug":"residual-connection","method_name":"Residual Connection"},{"method_slug":"softmax","method_name":"Softmax"},{"method_slug":"swin-transformer","method_name":"Swin Transformer"}],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/color-image-denoising-on-bsd300-lambda30","task":"Color Image Denoising","dataset":"BSD300 lambda30","model":"SwinIA","rank_in_archive_order":1,"of":1,"metrics":{"PSNR":"27.92","SSIM":"0.775"},"uses_additional_data":false},{"leaderboard":"/sota/color-image-denoising-on-bsd300-lambda5-50","task":"Color Image Denoising","dataset":"BSD300 lambda5-50","model":"SwinIA","rank_in_archive_order":1,"of":1,"metrics":{"PSNR":"27.74","SSIM":"0.764"},"uses_additional_data":false},{"leaderboard":"/sota/color-image-denoising-on-bsd300-sigma25","task":"Color Image Denoising","dataset":"BSD300 sigma25","model":"SwinIA","rank_in_archive_order":1,"of":1,"metrics":{"PSNR":"28.4","SSIM":"0.789"},"uses_additional_data":false},{"leaderboard":"/sota/color-image-denoising-on-bsd300-sigma5-50","task":"Color Image Denoising","dataset":"BSD300 sigma5-50","model":"SwinIA","rank_in_archive_order":1,"of":1,"metrics":{"PSNR":"28.4","SSIM":"0.785"},"uses_additional_data":false},{"leaderboard":"/sota/color-image-denoising-on-kodak24-lambda30","task":"Color Image Denoising","dataset":"Kodak24 lambda30","model":"SwinIA","rank_in_archive_order":1,"of":1,"metrics":{"PSNR":"29.51","SSIM":"0.805"},"uses_additional_data":false},{"leaderboard":"/sota/color-image-denoising-on-kodak24-lambda5-50","task":"Color Image Denoising","dataset":"Kodak24 lambda5-50","model":"SwinIA","rank_in_archive_order":1,"of":1,"metrics":{"PSNR":"29.06","SSIM":"0.788"},"uses_additional_data":false},{"leaderboard":"/sota/color-image-denoising-on-kodak24-sigma25","task":"Color Image Denoising","dataset":"Kodak24 sigma25","model":"SwinIA","rank_in_archive_order":3,"of":3,"metrics":{"PSNR":"30.12","SSIM":"0.819"},"uses_additional_data":false},{"leaderboard":"/sota/color-image-denoising-on-kodak24-sigma5-50","task":"Color Image Denoising","dataset":"Kodak24 sigma5-50","model":"SwinIA","rank_in_archive_order":1,"of":1,"metrics":{"PSNR":"30.3","SSIM":"0.820"},"uses_additional_data":false},{"leaderboard":"/sota/color-image-denoising-on-set14-lambda30","task":"Color Image Denoising","dataset":"Set14 lambda30","model":"SwinIA","rank_in_archive_order":1,"of":1,"metrics":{"PSNR":"28.74","SSIM":"0.799"},"uses_additional_data":false},{"leaderboard":"/sota/color-image-denoising-on-set14-lambda5-50","task":"Color Image Denoising","dataset":"Set14 lambda5-50","model":"SwinIA","rank_in_archive_order":1,"of":1,"metrics":{"PSNR":"28.27","SSIM":"0.780"},"uses_additional_data":false},{"leaderboard":"/sota/color-image-denoising-on-set14-sigma25","task":"Color Image Denoising","dataset":"Set14 sigma25","model":"SwinIA","rank_in_archive_order":1,"of":1,"metrics":{"PSNR":"29.54","SSIM":"0.814"},"uses_additional_data":false},{"leaderboard":"/sota/color-image-denoising-on-set14-sigma5-50","task":"Color Image Denoising","dataset":"Set14 sigma5-50","model":"SwinIA","rank_in_archive_order":1,"of":1,"metrics":{"PSNR":"29.49","SSIM":"0.809"},"uses_additional_data":false},{"leaderboard":"/sota/grayscale-image-denoising-on-bsd68-sigma15","task":"Grayscale Image Denoising","dataset":"BSD68 sigma15","model":"SwinIA","rank_in_archive_order":16,"of":16,"metrics":{"PSNR":"31.07","SSIM":"0.856"},"uses_additional_data":false},{"leaderboard":"/sota/grayscale-image-denoising-on-bsd68-sigma25","task":"Grayscale Image Denoising","dataset":"BSD68 sigma25","model":"SwinIA","rank_in_archive_order":12,"of":16,"metrics":{"PSNR":"29.17","SSIM":"0.801"},"uses_additional_data":false},{"leaderboard":"/sota/grayscale-image-denoising-on-bsd68-sigma50","task":"Grayscale Image Denoising","dataset":"BSD68 sigma50","model":"SwinIA","rank_in_archive_order":3,"of":15,"metrics":{"PSNR":"26.61","SSIM":"0.706"},"uses_additional_data":false},{"leaderboard":"/sota/grayscale-image-denoising-on-hanzi","task":"Grayscale Image Denoising","dataset":"Hanzi","model":"SwinIA","rank_in_archive_order":1,"of":1,"metrics":{"PSNR":"14.35","SSIM":"0.556"},"uses_additional_data":false},{"leaderboard":"/sota/grayscale-image-denoising-on-set12-sigma15","task":"Grayscale Image Denoising","dataset":"Set12 sigma15","model":"SwinIA","rank_in_archive_order":7,"of":8,"metrics":{"PSNR":"30.37","SSIM":"0.857"},"uses_additional_data":false},{"leaderboard":"/sota/grayscale-image-denoising-on-set12-sigma25","task":"Grayscale Image Denoising","dataset":"Set12 sigma25","model":"SwinIA","rank_in_archive_order":6,"of":6,"metrics":{"PSNR":"28.72","SSIM":"0.817"},"uses_additional_data":false},{"leaderboard":"/sota/grayscale-image-denoising-on-set12-sigma50","task":"Grayscale Image Denoising","dataset":"Set12 sigma50","model":"SwinIA","rank_in_archive_order":8,"of":8,"metrics":{"PSNR":"26.03","SSIM":"0.736"},"uses_additional_data":false},{"leaderboard":"/sota/medical-image-denoising-on-fmd-confocal-fish","task":"Medical Image Denoising","dataset":"FMD Confocal Fish","model":"SwinIA","rank_in_archive_order":1,"of":1,"metrics":{"PSNR":"31.79","SSIM":"0.871"},"uses_additional_data":false},{"leaderboard":"/sota/medical-image-denoising-on-fmd-confocal-mice","task":"Medical Image Denoising","dataset":"FMD Confocal Mice","model":"SwinIA","rank_in_archive_order":1,"of":1,"metrics":{"PSNR":"37.65","SSIM":"0.960"},"uses_additional_data":false},{"leaderboard":"/sota/medical-image-denoising-on-fmd-two-photon","task":"Medical Image Denoising","dataset":"FMD Two-Photon Mice","model":"SwinIA","rank_in_archive_order":1,"of":1,"metrics":{"PSNR":"33.25","SSIM":"0.915"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=2305.05651","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}