Woven fab­rics have a wide range of ap­pear­ance de­term­ined by their small-scale 3D struc­ture. Ac­cur­ately mod­el­ing this struc­tur­al de­tail can pro­duce highly real­ist­ic ren­der­ings of fab­rics and is crit­ic­al for pre­dict­ive ren­der­ing of fab­ric ap­pear­ance. But build­ing these yarn-level volu­met­ric mod­els is chal­len­ging. Pro­ced­ur­al tech­niques are manu­ally in­tens­ive, and fail to cap­ture the nat­ur­ally arising ir­reg­u­lar­it­ies which con­trib­ute sig­ni­fic­antly to the over­all ap­pear­ance of cloth. Tech­niques that ac­quire the de­tailed 3D struc­ture of real fab­ric samples are con­strained only to mod­el the scanned samples and can­not rep­res­ent dif­fer­ent fab­ric designs.

This pa­per presents a new ap­proach to cre­at­ing volu­met­ric mod­els of woven cloth, which starts with user-spe­cified fab­ric designs and pro­duces mod­els that cor­rectly cap­ture the yarn-level struc­tur­al de­tails of cloth. We cre­ate a small data­base of volu­met­ric ex­em­plars by scan­ning fab­ric samples with simple weave struc­tures. To build an out­put mod­el, our meth­od syn­thes­izes a new volume by copy­ing data from the ex­em­plars at each yarn cross­ing to match a weave pat­tern that spe­cifies the de­sired out­put struc­ture. Our res­ults demon­strate that our ap­proach gen­er­al­izes well to com­plex designs and can pro­duce highly real­ist­ic res­ults at both large and small scales.

Pa­per fig­ure se­lec­ted as the front cov­er of the SIGGRAPH 2012 con­fer­ence pro­ceed­ings