{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,5,4]],"date-time":"2026-05-04T04:47:03Z","timestamp":1777870023701,"version":"3.51.4"},"reference-count":31,"publisher":"Wiley","license":[{"start":{"date-parts":[[2026,4,30]],"date-time":"2026-04-30T00:00:00Z","timestamp":1777507200000},"content-version":"vor","delay-in-days":0,"URL":"http:\/\/creativecommons.org\/licenses\/by\/4.0\/"},{"start":{"date-parts":[[2026,4,30]],"date-time":"2026-04-30T00:00:00Z","timestamp":1777507200000},"content-version":"tdm","delay-in-days":0,"URL":"http:\/\/doi.wiley.com\/10.1002\/tdm_license_1.1"}],"content-domain":{"domain":["onlinelibrary.wiley.com"],"crossmark-restriction":true},"short-container-title":["Computer Graphics Forum"],"abstract":"<jats:title>Abstract<\/jats:title>\n                  <jats:p>Classical diffusion models typically rely on isotropic Gaussian noise, treating all regions uniformly and overlooking structural information important for high\u2010quality generation. We introduce an edge\u2010preserving diffusion process that generalizes isotropic models via a hybrid noise scheme with an edge\u2010aware scheduler that smoothly transitions from edge\u2010preserving to isotropic noise. This enables the model to capture fine structural details while generally maintaining global performance. We evaluate the impact of structure\u2010aware noise in both diffusion and flow\u2010matching frameworks, and show that existing isotropic models can be efficiently fine\u2010tuned with edge\u2010preserving noise, making our framework practical for adapting pre\u2010trained systems. Beyond unconditional generation, our method particularly shows improvements in structure\u2010guided tasks such as stroke\u2010to\u2010image synthesis, improving robustness and perceptual quality, as evidenced by consistent improvements across FID, KID, and CLIP\u2010score.<\/jats:p>","DOI":"10.1111\/cgf.70383","type":"journal-article","created":{"date-parts":[[2026,4,30]],"date-time":"2026-04-30T15:12:50Z","timestamp":1777561970000},"update-policy":"https:\/\/doi.org\/10.1002\/crossmark_policy","source":"Crossref","is-referenced-by-count":0,"title":["Edge\u2010preserving noise for diffusion models"],"prefix":"10.1111","author":[{"ORCID":"https:\/\/orcid.org\/0000-0002-9488-9327","authenticated-orcid":false,"given":"Jente","family":"Vandersanden","sequence":"first","affiliation":[{"name":"Max Planck Institute for Informatics  Germany"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"ORCID":"https:\/\/orcid.org\/0009-0004-7245-7998","authenticated-orcid":false,"given":"Sascha","family":"Holl","sequence":"additional","affiliation":[{"name":"Max Planck Institute for Informatics  Germany"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-2769-8408","authenticated-orcid":false,"given":"Xingchang","family":"Huang","sequence":"additional","affiliation":[{"name":"Max Planck Institute for Informatics  Germany"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-0970-5835","authenticated-orcid":false,"given":"Gurprit","family":"Singh","sequence":"additional","affiliation":[{"name":"Max Planck Institute for Informatics  Germany"},{"name":"Advanced Micro Devices  Germany"}],"role":[{"role":"author","vocabulary":"crossref"}]}],"member":"311","published-online":{"date-parts":[[2026,4,30]]},"reference":[{"key":"e_1_2_9_2_2","doi-asserted-by":"crossref","DOI":"10.52202\/075280-1789","article-title":"Cold diffusion: Inverting arbitrary image transforms without noise","volume":"36","author":"Bansal A.","year":"2023","journal-title":"Advances in Neural Information Processing Systems"},{"key":"e_1_2_9_3_2","doi-asserted-by":"crossref","unstructured":"ChoiY. UhY. YooJ. HaJ.-W.: Stargan v2: Diverse image synthesis for multiple domains. InProceedings of the IEEE\/CVF conference on computer vision and pattern recognition(2020) pp.8188\u20138197. 10","DOI":"10.1109\/CVPR42600.2020.00821"},{"key":"e_1_2_9_4_2","unstructured":"DarasG. DelbracioM. TalebiH. DlmakisA. MilanfarP.: Soft diffusion: Score matching with general corruptions.Transactions on Machine Learning Research(2023). 2"},{"key":"e_1_2_9_5_2","unstructured":"DockhornT. VahdatA. KreisK.: Score-based generative modeling with critically-damped langevin diffusion. InInternational Conference on Learning Representations(2022). 2"},{"issue":"4","key":"e_1_2_9_6_2","doi-asserted-by":"crossref","first-page":"44:1","DOI":"10.1145\/2185520.2185540","article-title":"How do humans sketch objects?","volume":"31","author":"Eitz M.","year":"2012","journal-title":"ACM Trans. Graph. (Proc. SIGGRAPH)"},{"issue":"3","key":"e_1_2_9_7_2","doi-asserted-by":"crossref","first-page":"1594","DOI":"10.1137\/23M1545859","article-title":"Image denoising: The deep learning revolution and beyond\u2014a survey paper","volume":"16","author":"Elad M.","year":"2023","journal-title":"SIAM Journal on Imaging Sciences"},{"key":"e_1_2_9_8_2","first-page":"6840","article-title":"Denoising diffusion probabilistic models","volume":"33","author":"Ho J.","year":"2020","journal-title":"Advances in neural information processing systems"},{"key":"e_1_2_9_9_2","article-title":"Gans trained by a two time-scale update rule converge to a local nash equilibrium","volume":"30","author":"Heusel M.","year":"2017","journal-title":"Advances in neural information processing systems"},{"key":"e_1_2_9_10_2","unstructured":"HoogeboomE. SalimansT.: Blurring diffusion models. InThe Eleventh International Conference on Learning Representations(2023). 2"},{"key":"e_1_2_9_11_2","doi-asserted-by":"crossref","unstructured":"HuangX. SalaunC. VasconcelosC. TheobaltC. OztireliC. SinghG.: Blue noise for diffusion models. InACM SIGGRAPH 2024 Conference Papers(2024) pp.1\u201311. 2 6","DOI":"10.1145\/3641519.3657435"},{"key":"e_1_2_9_12_2","unstructured":"KingmaD. P. BaJ.: Adam: A method for stochastic optimization.arXiv preprint arXiv:1412.6980(2014). 10"},{"key":"e_1_2_9_13_2","unstructured":"KrizhevskyA. HintonG. et al.:Learning multiple layers of features from tiny images. 10"},{"key":"e_1_2_9_14_2","article-title":"Improved precision and recall metric for assessing generative models","volume":"32","author":"Kynk\u00e4\u00e4nniemi T.","year":"2019","journal-title":"Advances in neural information processing systems"},{"key":"e_1_2_9_15_2","first-page":"21696","article-title":"Variational diffusion models","volume":"34","author":"Kingma D.","year":"2021","journal-title":"Advances in neural information processing systems"},{"key":"e_1_2_9_16_2","unstructured":"LipmanY. ChenR. T. Ben-HamuH. NickelM. LeM.: Flow matching for generative modeling.arXiv preprint arXiv:2210.02747(2022). 5 6 7 10"},{"key":"e_1_2_9_17_2","doi-asserted-by":"crossref","unstructured":"LeeC.-H. LiuZ. WuL. LuoP.: Maskgan: Towards diverse and interactive facial image manipulation. InIEEE Conference on Computer Vision and Pattern Recognition (CVPR)(2020). 10","DOI":"10.1109\/CVPR42600.2020.00559"},{"key":"e_1_2_9_18_2","unstructured":"MengC. HeY. SongY. SongJ. WuJ. ZhuJ.-Y. ErmonS.: SDEdit: Guided image synthesis and editing with stochastic differential equations. InInternational Conference on Learning Representations(2022). 1 6 11 12 13"},{"key":"e_1_2_9_19_2","unstructured":"PaszkeA. GrossS. ChintalaS. ChananG. YangE. DeVitoZ. LinZ. DesmaisonA. AntigaL. LererA.:Automatic differentiation in pytorch. 10"},{"issue":"7","key":"e_1_2_9_20_2","doi-asserted-by":"crossref","first-page":"629","DOI":"10.1109\/34.56205","article-title":"Scale-space and edge detection using anisotropic diffusion","volume":"12","author":"Perona P.","year":"1990","journal-title":"IEEE Transactions on pattern analysis and machine intelligence"},{"key":"e_1_2_9_21_2","unstructured":"RombachR. BlattmannA. LorenzD. EsserP. OmmerB.: High-resolution image synthesis with latent diffusion models. InProceedings of the IEEE\/CVF conference on computer vision and pattern recognition(2022) pp.10684\u201310695. 6 10"},{"key":"e_1_2_9_22_2","unstructured":"RissanenS. HeinonenM. SolinA.: Generative modelling with inverse heat dissipation. InThe Eleventh International Conference on Learning Representations(2023). 2 10"},{"key":"e_1_2_9_23_2","first-page":"8748","volume-title":"International conference on machine learning","author":"Radford A.","year":"2021"},{"key":"e_1_2_9_24_2","article-title":"Exposing flaws of generative model evaluation metrics and their unfair treatment of diffusion models","volume":"36","author":"Stein G.","year":"2024","journal-title":"Advances in Neural Information Processing Systems"},{"key":"e_1_2_9_25_2","first-page":"2256","volume-title":"International conference on machine learning","author":"Sohl-Dickstein J.","year":"2015"},{"key":"e_1_2_9_26_2","article-title":"Generative modeling by estimating gradients of the data distribution","volume":"32","author":"Song Y.","year":"2019","journal-title":"Advances in neural information processing systems"},{"key":"e_1_2_9_27_2","unstructured":"SongY. Sohl-DicksteinJ. KingmaD. P. KumarA. ErmonS. PooleB.: Score-based generative modeling through stochastic differential equations. InInternational Conference on Learning Representations(2021). 1 2"},{"key":"e_1_2_9_28_2","doi-asserted-by":"crossref","unstructured":"SzegedyC. VanhouckeV. IoffeS. ShlensJ. WojnaZ.: Rethinking the inception architecture for computer vision. InProceedings of the IEEE conference on computer vision and pattern recognition(2016) pp.2818\u20132826. 10","DOI":"10.1109\/CVPR.2016.308"},{"key":"e_1_2_9_29_2","unstructured":"VoletiV. PalC. ObermanA. M.: Score-based denoising diffusion with non-isotropic gaussian noise models. InNeurIPS 2022 Workshop on Score-Based Methods(2022). 2"},{"key":"e_1_2_9_30_2","volume":"1","author":"Weickert J.","year":"1998","journal-title":"Anisotropic diffusion in image processing"},{"key":"e_1_2_9_31_2","article-title":"Constructing non-isotropic gaussian diffusion model using isotropic gaussian diffusion model for image editing","volume":"36","author":"Yu X.","year":"2024","journal-title":"Advances in Neural Information Processing Systems"},{"key":"e_1_2_9_32_2","unstructured":"YuF. SeffA. ZhangY. SongS. FunkhouserT. XiaoJ.: Lsun: Constructionof a large-scale image dataset using deep learning with humans in the loop.arXiv preprint arXiv:1506.03365(2015). 10"}],"container-title":["Computer Graphics Forum"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/onlinelibrary.wiley.com\/doi\/pdf\/10.1111\/cgf.70383","content-type":"application\/pdf","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/onlinelibrary.wiley.com\/doi\/full-xml\/10.1111\/cgf.70383","content-type":"application\/xml","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/onlinelibrary.wiley.com\/doi\/pdf\/10.1111\/cgf.70383","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,4,30]],"date-time":"2026-04-30T15:12:57Z","timestamp":1777561977000},"score":1,"resource":{"primary":{"URL":"https:\/\/onlinelibrary.wiley.com\/doi\/10.1111\/cgf.70383"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,4,30]]},"references-count":31,"alternative-id":["10.1111\/cgf.70383"],"URL":"https:\/\/doi.org\/10.1111\/cgf.70383","archive":["Portico"],"relation":{},"ISSN":["0167-7055","1467-8659"],"issn-type":[{"value":"0167-7055","type":"print"},{"value":"1467-8659","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,4,30]]},"assertion":[{"value":"2026-04-30","order":3,"name":"published","label":"Published","group":{"name":"publication_history","label":"Publication History"}}],"article-number":"e70383"}}