BEGIN:VCALENDAR VERSION:2.0 PRODID:Linklings LLC BEGIN:VTIMEZONE TZID:Asia/Seoul X-LIC-LOCATION:Asia/Seoul BEGIN:STANDARD TZOFFSETFROM:+0900 TZOFFSETTO:+0900 TZNAME:KST DTSTART:18871231T000000 DTSTART:19881009T020000 END:STANDARD END:VTIMEZONE BEGIN:VEVENT DTSTAMP:20230103T035312Z LOCATION:Room 324\, Level 3\, West Wing DTSTART;TZID=Asia/Seoul:20221208T170000 DTEND;TZID=Asia/Seoul:20221208T183000 UID:siggraphasia_SIGGRAPH Asia 2022_sess167_papers_361@linklings.com SUMMARY:MIPNet: Neural Normal-to-Anisotropic-Roughness MIP mapping DESCRIPTION:Technical Communications, Technical Papers\n\nMIPNet: Neural N ormal-to-Anisotropic-Roughness MIP mapping\n\nGauthier, Faury, Levallois, Thonat, Thiery...\n\nWe present MIPNet, a novel approach for SVBRDF mipmap ping which preserves material appearance under varying view distances and lighting conditions. As in classical mipmapping, our method explicitly enc odes the multiscale appearance of materials in a SVBRDF mipmap pyramid. To do so, we use a tensor-based representation, coping with gradient-based o ptimization, for encoding anisotropy which is compatible with existing rea l-time rendering engines. Instead of relying on a simple texture patch ave rage for each channel independently, we propose a cascaded architecture of multilayer perceptrons to approximate the material appearance using only the fixed material channels. Our neural model learns simple mipmapping fil ters using a differentiable rendering pipeline based on a rendering loss a nd is able to transfer signal from normal to anisotropic roughness. As a r esult, we obtain a drop-in replacement for standard material mipmapping, o ffering a significant improvement in appearance preservation while still b oiling down to a single per-pixel mipmap texture fetch. We report extensiv e experiments on two distinct BRDF models.\n\nRegistration Category: FULL ACCESS, ON-DEMAND ACCESS\n\nLanguage: ENGLISH\n\nFormat: IN-PERSON, ON-DEM AND URL:https://sa2022.siggraph.org/en/full-program/?id=papers_361&sess=sess16 7 END:VEVENT END:VCALENDAR