Real­ist­ic mod­el­ing of the bi­d­irec­tion­al re­flect­ance dis­tri­bu­tion func­tion (BRDF) of scene ob­jects is a vi­tal pre­requis­ite for any type of phys­ic­ally based ren­der­ing. In the last dec­ades, the avail­ab­il­ity of data­bases con­tain­ing real-world ma­ter­i­al meas­ure­ments has fueled con­sid­er­able in­nov­a­tion in the de­vel­op­ment of such mod­els. However, pre­vi­ous work in this area was mainly fo­cused on in­creas­ing the visu­al real­ism of im­ages, and hence ig­nored the ef­fect of scat­ter­ing on the po­lar­iz­a­tion state of light, which is nor­mally im­per­cept­ible to the hu­man eye. Ex­ist­ing data­bases thus only cap­ture scattered flux, or they are too dir­ec­tion­ally sparse (e.g., in-plane) to be us­able for sim­u­la­tion.

While subtle to hu­man ob­serv­ers, po­lar­iz­a­tion is eas­ily per­ceived by any op­tic­al sensor (e.g., us­ing po­lar­iz­ing fil­ters), provid­ing a wealth of ad­di­tion­al in­form­a­tion about shape and ma­ter­i­al prop­er­ties of the ob­ject un­der ob­ser­va­tion. Giv­en the in­creas­ing ap­plic­a­tion of ren­der­ing in the solu­tion of in­verse prob­lems via ana­lys­is-by-syn­thes­is and dif­fer­en­ti­ation, the abil­ity to real­ist­ic­ally mod­el po­lar­ized ra­di­at­ive trans­port is thus highly de­sir­able.

Po­lar­iz­a­tion de­pends on the wavelength of the spec­trum, and thus we provide the first po­lar­i­met­ric BRDF (pBRDF) data­set that cap­tures the po­lar­i­met­ric prop­er­ties of real-world ma­ter­i­als over the full an­gu­lar do­main, and at mul­tiple wavelengths. Ac­quis­i­tion of such re­flect­ance data is chal­len­ging due to the ex­tremely large space of an­gu­lar, spec­tral, and po­lar­i­met­ric con­fig­ur­a­tions that must be ob­served, and we pro­pose a scheme com­bin­ing im­age-based ac­quis­i­tion with spec­tro­scop­ic el­lip­so­metry to per­form meas­ure­ments in a real­ist­ic amount of time. This pro­cess yields raw Mueller matrices, which we sub­sequently trans­form in­to Ru­sinkiewicz-para­met­er­ized pBRD­Fs that can be used for ren­der­ing.

Our data­set provides 25 iso­trop­ic pBRD­Fs span­ning a wide range of ap­pear­ances: dif­fuse/spec­u­lar, metal­lic/dielec­tric, rough/smooth, and dif­fer­ent col­or al­bedos, cap­tured in five wavelength ranges cov­er­ing the vis­ible spec­trum. We demon­strate us­age of our data-driv­en pBRDF mod­el in a phys­ic­ally based ren­der­er that ac­counts for po­lar­ized in­ter­re­flec­tion, and we in­vest­ig­ate the re­la­tion­ship of po­lar­iz­a­tion and ma­ter­i­al ap­pear­ance, provid­ing in­sights in­to the be­ha­vi­or of char­ac­ter­ist­ic real-world pBRD­Fs.