• Material Measurement Laboratory /
• Material database

Ma­ter­i­al Data­base — Fre­quently Asked Ques­tions

Fea­tures

The ma­ter­i­al data­base provides broad cov­er­age of com­mon and more spe­cial­ized ma­ter­i­als, in­clud­ing iso­trop­ic and an­iso­trop­ic BRD­Fs of metals, pa­per, car paints, or­gan­ic samples, fab­rics, etc. Each ma­ter­i­al is avail­able in a spec­tral ver­sion that cov­ers the 360(UV) - 1000nm (NIR) range with ~4nm sample spa­cing, as well as a RGB ver­sion for com­pat­ib­il­ity with ren­der­ers that do not sup­port col­or spec­tra. The rep­res­ent­a­tion is ex­tremely com­pact, re­quir­ing ap­prox­im­ately 16KiB and 544KiB per chan­nel for iso­trop­ic and an­iso­trop­ic ma­ter­i­als, re­spect­ively. Fur­ther­more, it provides a nat­ur­al im­port­ance sampling op­er­a­tion that does not re­quire nu­mer­ic­al fits or com­plex ad­di­tion­al data struc­tures. Tech­nic­al de­tails on the un­der­ly­ing para­met­er­iz­a­tion and meas­ure­ment meth­od­o­logy are avail­able in the pa­per

An Ad­apt­ive Para­met­er­iz­a­tion for Ef­fi­cient Ma­ter­i­al Ac­quis­i­tion and Ren­der­ing
by Jonath­an Dupuy and Wen­zel Jakob
In ACM Trans­ac­tions on Graph­ics (Pro­ceed­ings of SIGGRAPH Asia 2018)

The meas­ure­ment device and its cap­ab­il­it­ies are de­scribed here.

Fu­ture plans

We plan to con­sid­er­ably ex­tend the data­base in the fu­ture so that it even­tu­ally con­tains hun­dreds of ma­ter­i­als that densely cov­er the space of real-world ma­ter­i­al ap­pear­ance. Out­side help is wel­come: if you have in­ter­est­ing samples, we'd love to meas­ure them. See be­low for de­tails.

Stan­dalone view­er

BRDF files can be in­ter­act­ively viewed and ana­lyzed us­ing the Tekari view­er cre­ated by Ben­oît Ruiz. Tekari plots the BRDF and sample loc­a­tions with re­spect to dif­fer­ent in­cid­ent angles and/or wavelengths, us­ing a vari­ety of power­ful visu­al­iz­a­tion tools. Click the fol­low­ing link to down­load pre­com­piled 64-bit ex­ecut­ables for OSX, Win­dows, and Linux. To open a ma­ter­i­al, simply drag & drop the “_spec.bsdf” file onto the ap­plic­a­tion, or provide the fi­le­name as a com­mand line ar­gu­ment. Note while Tekari is an open source pro­ject, it de­pends on a com­pon­ent, whose source code is not yet avail­able (hence the pre­com­piled bin­ar­ies). 

C++ and Py­thon API

We've re­leased a ref­er­ence BRDF im­ple­ment­a­tion, which provides the stand­ard op­er­a­tions needed by most mod­ern ren­der­ing sys­tems: BRDF eval­u­ation, sampling, and a routine to query the un­der­ly­ing prob­ab­il­ity dens­ity. The re­pos­it­ory also con­tains a plu­gin for the Mit­suba ren­der­er and Py­thon code to load the data files in­to NumPy, which is use­ful for visu­al­iz­a­tion and sys­tem­at­ic ana­lys­is of the data­base con­tents.

Raw data

A go­nio-pho­to­met­er typ­ic­ally can­not meas­ure all angle com­bin­a­tions due to op­tic­al and mech­an­ic­al lim­it­a­tions (for in­stance, the sensor oc­cludes the light source at cer­tain angles). Our meas­ure­ment pipeline de­tects such cases and ex­tra­pol­ates the miss­ing data from oth­er meas­ure­ments. To provide trans­par­ency on what parts of the data cor­res­pond to ac­tu­al meas­ure­ments, we provide “raw” data files con­tain­ing an ex­tremely verb­ose log of the en­tire ac­quis­i­tion pro­cess in Py­thon pickle format.

Un­der­stand­ing the visu­al­iz­a­tions

Click­ing on the ma­ter­i­al re­veals a num­ber of visu­al­iz­a­tions:

Slice (meas­ured): The first im­age shows an RGB rendi­tion of what was meas­ured by the spec­tro­met­er. Note that the data typ­icaly has ex­tremely high-dy­nam­ic range, hence some de­tail in the black may simply be in­vis­ible. Red pixels cor­res­pond to wasted space of the dis­cret­iz­a­tion. The black column on the left is a re­gion that could not be meas­ured.

Slice (pro­cessed): This plot con­tains the same data, but after post-pro­cessing. The black re­gions on the left were filled in, and the data was scaled by the Jac­obi­an of the para­met­er­iz­a­tion (see the pa­per for de­tails). This has the ef­fect of mak­ing the rep­res­ent­a­tion sig­ni­fic­antly easi­er to in­ter­pol­ate. The bot­tom row shows spec­tral curves in the 360..1000nm range cor­res­pond­ing to the row above.

Para­met­er­iz­a­tion: shows in­cid­ent and out­go­ing angles, for which the ma­ter­i­al was meas­ured.

NDF: Our meth­od heav­ily re­lies on the ma­ter­i­al's retrore­flect­ive re­ponse. This plot shows the ex­trac­ted nor­mal dis­tri­bu­tion and the res­ult­ing backs­cat­ter­ing (the­ory & ac­tu­al).

Meas­ured angles: 2D visu­al­iz­a­tion of the meas­ured el­ev­a­tions (left) and azi­muths (right)

Dis­cret­iz­a­tion/noise ar­ti­facts in view­er?

A num­ber of ma­ter­i­als may show grid/noise-like ar­ti­facts when visu­al­ized in the Tekari view­er. This is not a prob­lem of the data, but rather an ar­ti­fact of the visu­al­iz­a­tion meth­od in­volving Delaunay tri­an­gu­la­tion. You can simply in­crease the res­ol­u­tion in the “Sampling Res­ol­u­tion” pan­el.

Li­cens­ing con­di­tions

Un­less oth­er­wise noted, all ma­ter­i­al data is li­censed un­der the Cre­at­ive Com­mons Zero (CC0) li­cense.

Meas­ure­ment ser­vice

We plan to con­sid­er­ably ex­tend the data­base in the fu­ture and would be happy to meas­ure samples provided by you for free. Please reach out to us if this sounds in­ter­est­ing. There are a few strings at­tached:

• We are mainly in­ter­ested in ma­ter­i­als that are suf­fi­ciently in­ter­est­ing/dif­fer­ent from ones that are already in the data­base, so do reach out be­fore mail­ing us something.
• It must be okay for us to in­clude the res­ult­ing meas­ure­ment in the data­base un­der a CC0 li­cense.
• Samples must be flat and suf­fi­ciently large (>= 10x10cm, ideally >= 20x20cm) so that they can be ac­com­mod­ated by our sample hold­er. If you have odd-sized, non-rect­an­gu­lar, and/or non-flat samples, please send them to us moun­ted on a suit­able rect­an­gu­lar sub­strate.
• We con­stantly im­prove our meas­ure­ment pipeline and will oc­ca­sion­ally re-meas­ure large sets of samples when the pro­cess has changed bey­ond a cer­tain point. To be able to do this, we need to phys­ic­ally keep the samples at our lab in Lausanne, Switzer­land. We are only able to re­turn-samples if there are ex­tremely good reas­ons to do so (very ex­pens­ive samples like pre­cious metals, unique spe­ci­mens, etc.). In this case, we would ask you to cov­er any re­turn mail­ing costs in­clud­ing in­sur­ance.

Changelog

We routinely add ad­di­tion­al ma­ter­i­als to the data­base. More sig­ni­fic­ant changes will be lis­ted here.

• Nov 1, 2018. A lar­ger stur­di­er sensor arm (which can be used to mount high-res ca­mas) was in­stalled on the go­nio-pho­to­met­er. One po­ten­tial down­side of the change is that the head is lar­ger and thus in­creases the set of shad­owed con­fig­ur­a­tion where the head par­tially or fully blocks the col­lim­ated beam. It is worth not­ing that the com­pu­ta­tion used to de­tect such shad­owed con­fig­ur­a­tions is some­what con­ser­vat­ive. To en­able fu­ture im­prove­ments in case an im­proved shad­ow­ing de­tec­tion lo­gic is de­veloped later, RAW files pro­duced by the latest meas­ure­ment pipeline (v3.3+) also re­cord meas­ure­ments even for sup­posedly shad­owed con­fig­ur­a­tions.
• Nov 14, 2018. Re­meas­ured & pro­cessed all silk_* and vch_* ma­ter­i­als with the latest meas­ure­ment pipeline. The pre­vi­ous data­set con­tained vis­ible in­ter­pol­a­tion ar­ti­facts pro­duced by an older pipeline.
• Nov 20, 2018: We've ad­ded a num­ber of pa­per samples of dif­fer­ent col­ors. Pa­per, es­pe­cially the white vari­ant, con­tains bright­en­ing ad­dit­ives, that ab­sorb in the UV and emit in the blue range. Our cur­rent meas­ure­ments setup does not ac­count for such fluor­es­cent ef­fects, which would re­quire sep­ar­ate meas­ure­ment of the in­cid­ent and out­go­ing wavelengths. As a res­ult, the ma­ter­i­al is seen to emit more light than it re­ceives, which breaks en­ergy con­ser­va­tion and pro­duces con­ver­gence ar­ti­facts in ren­der­ings. We still in­clude such ma­ter­i­als in the data­base, but they are likely not dir­ectly us­able.