{"id":41061,"date":"2025-07-01T18:01:44","date_gmt":"2025-07-01T18:01:44","guid":{"rendered":"https:\/\/www.labri.fr\/perso\/barla\/blog\/?p=41061"},"modified":"2025-07-01T18:01:45","modified_gmt":"2025-07-01T18:01:45","slug":"egsr-2025","status":"publish","type":"post","link":"https:\/\/www.labri.fr\/perso\/barla\/blog\/?p=41061","title":{"rendered":"EGSR 2025"},"content":{"rendered":"<p id=\"top\" \/>\n<p>Weidlich keynote:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Layered materials including defects<\/li>\n\n\n\n<li>No complex vertical layering most of the time (except for thick layers)<\/li>\n\n\n\n<li>Classical PBR materials are too limited, especially for grazing angle effects<\/li>\n\n\n\n<li>Complex Layered materials are too expensive to store\/render : use neural approach, with quality control.<\/li>\n\n\n\n<li>Analytic materials &amp; rendering remain important for controllability and consistency in production.<\/li>\n<\/ul>\n\n\n\n<p>RT rendering session:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Two papers on hair: a strand model (check shadow-masking term!) and a g-buffer based denoiser<\/li>\n\n\n\n<li>Image-based Rendering of glints: avoid sampling envmap for all discrete facet directions, by counting them inside areas of intensity-segmented envmap<\/li>\n<\/ul>\n\n\n\n<p>Sampling &amp; guiding:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Dwivedi guiding improvement for rendering translucent materials via caching statistics<\/li>\n<\/ul>\n\n\n\n<p>Yu keynote:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Startup-like videos with ukulele, full of labyrinthine AI pipelines and off-the-mark citations (like Feynmann&#8217;s about understanding)<\/li>\n<\/ul>\n\n\n\n<p>Light &amp; brightness session:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Complex luminaires through a precomputed prefiltered light field over convex hull, stored and importance sampled via a wavelet decomposition.<\/li>\n\n\n\n<li>Multiscale divisive normalization model captures (average) brightness perception while preserving sharpness; use in local tone mapping operator, but requires handling of dark halos.<\/li>\n<\/ul>\n\n\n\n<p>Gaussians session:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Build panorama from GS scene through dedicated raytracing of Gaussians.<\/li>\n\n\n\n<li>2DGS with textures in 2d space aligned with gaussian plane to capture fine texture details. But view-dependency is still dealt with through SH.<\/li>\n\n\n\n<li>Stochastic ray-gaussian intersection through binarization of gaussian opacities. BVH helps by roughly sorting gaussians front to back along ray.<\/li>\n<\/ul>\n\n\n\n<p>Stylization session:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Create a database of stylized depictions using StylIt to provide text-based generation in a specific format.<\/li>\n<\/ul>\n\n\n\n<p>Stork keynote:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Use several estimates of light direction (based on contours, shadows, highlights) in paintings, and through Maximum likelihood integration of estimates evaluate their coherence.<\/li>\n\n\n\n<li>Reconstruct 3d scenes of paintings to check claims of Hockney that some painters used optics to get some effects (perspective, reflections) right.<\/li>\n<\/ul>\n\n\n\n<p>Material session:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Corona effect: avoid precomputation of diffraction pattern by expressing diffraction texture as a stochastic distribution of scattered and relying on multiplication in Fourier space, and combine with structure factor to account for correlations. P: Fraunhoffer assumption of far field still valid in case of water droplets on glasses?<\/li>\n\n\n\n<li>Liquid foams: cross section of plateau borders between bubbles treated like hairs or fur; using NN for multiple scattering. P: variation of liquid amount in borders due to gravity affects both foam shape and appearance!<\/li>\n<\/ul>\n\n\n\n<p>Analytical VS learned:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Tobias: agree to loose grasp on understanding, for the benefit of sharing a common methodology across problems\/disciplines<\/li>\n\n\n\n<li>But: only superficial sharing, same as globish in human communication, potentially leading to an impoverishment of thoughts (c.f. law of the instrument?)<\/li>\n\n\n\n<li>Still: requires a reference simulation to build analytical or learned solutions anyways, and there is also value in trying to solve the same problem with different solutions (and most likely different approximations as well)<\/li>\n\n\n\n<li>In practice: deep learning cannot guarantee the accuracy of results, cannot easily explain the nature of its approximations, and hyper-parameter tuning might reduce potential for sharing across problems<\/li>\n\n\n\n<li>Now: in which case one solution or the other is warranted? Would there be intermediate solutions that retain some (higher) level of physical understanding? Is the divide simply a science VS engineering one? Should we organize a panel discussion on the topic @ EGSR?<\/li>\n<\/ul>\n\n\n\n<p>Symbolic MC:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Could be used for a symbolic virtual gonio, but requires a stochastic description of microgeometry for interfaces and media should they act as symbolic variables<\/li>\n\n\n\n<li>First apply to complex microstructures (layered media, micro-shell surfaces) to model complex multiple scattering BSDF component functions of material properties of interest. Store polynomial coefficients in binned hemispheres. Use a scaling approach to vary global thickness?<\/li>\n\n\n\n<li>Then apply to macroscopic shapes filled with uniform media to model the appearance through shape-dependent BSSRDFs. But this will greatly increase the dimensionality of the function, and will likely require distant viewing and\/or lighting assumptions. Again, use a scaling approach to vary global scale?<\/li>\n\n\n\n<li>Finally, apply through a photon map approach, for instance to Compute luminaire functions, that many finally be final gathered through (differential) path tracing.<\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"<p>Weidlich keynote: RT rendering session: Sampling &amp; guiding: Yu keynote: Light &amp; brightness session: Gaussians session: Stylization session: Stork keynote: Material session: Analytical VS learned: Symbolic MC:<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[81],"tags":[],"class_list":["post-41061","post","type-post","status-publish","format-standard","hentry","category-talks"],"_links":{"self":[{"href":"https:\/\/www.labri.fr\/perso\/barla\/blog\/index.php?rest_route=\/wp\/v2\/posts\/41061","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.labri.fr\/perso\/barla\/blog\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.labri.fr\/perso\/barla\/blog\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.labri.fr\/perso\/barla\/blog\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.labri.fr\/perso\/barla\/blog\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=41061"}],"version-history":[{"count":1,"href":"https:\/\/www.labri.fr\/perso\/barla\/blog\/index.php?rest_route=\/wp\/v2\/posts\/41061\/revisions"}],"predecessor-version":[{"id":41062,"href":"https:\/\/www.labri.fr\/perso\/barla\/blog\/index.php?rest_route=\/wp\/v2\/posts\/41061\/revisions\/41062"}],"wp:attachment":[{"href":"https:\/\/www.labri.fr\/perso\/barla\/blog\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=41061"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.labri.fr\/perso\/barla\/blog\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=41061"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.labri.fr\/perso\/barla\/blog\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=41061"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}