RGL
Realistic Graphics Lab
EPFL Logo

Open Po­s­i­tions

RGL is look­ing to re­cruit a postdoc and a PhD stu­dent in the wider area of dif­fer­en­ti­able and in­verse ren­der­ing / ren­der­ing sys­tems / ap­pear­ance mod­el­ing in the up­com­ing ad­mis­sions cycle (Dead­line for PhD ap­plic­ants: Decem­ber 15, 2020. Postdocs: no dead­line, just drop me a mes­sage).

Are you ex­cited about any of the fol­low­ing top­ics?

Dif­fer­en­ti­able ren­der­ing

In ad­di­tion to gen­er­at­ing an im­age, a dif­fer­en­ti­able ren­der­ing al­gorithm propag­ates de­riv­at­ive in­form­a­tion through the en­tire sim­u­la­tion (a bit like Py­T­orch/Tensor­Flow, but highly spe­cial­ized to the phys­ics of light). This en­ables fas­cin­at­ing and com­pletely new uses of ren­der­ing al­gorithms to solve in­verse prob­lems. RGL is de­vel­op­ing Mit­suba 2, a new ren­der­ing sys­tem that can be used to this end. Please take a look at the fol­low­ing three RGL pro­jects to learn about our work in this area: [1] [2] [3].

 

Ren­der­ing sys­tems

Mod­ern ren­der­ing sys­tems are large and com­plex. They in­volve ad­vanced phys­ic­al ef­fects (spec­tra, po­lar­iz­a­tion, wave-op­tic­al in­ter­fer­ence), GPU and CPU im­ple­ment­a­tions, dif­fer­en­ti­ation, and a vari­ety of dif­fer­ent ren­der­ing tech­niques (Bi­d­irec­tion­al Path Tra­cing, Met­ro­pol­is Light Trans­port, Ver­tex Con­nec­tion Mer­ging, Gradi­ent Do­main meth­ods, etc.). Deal­ing with all of this com­plex­ity us­ing stand­ard pro­gram­ming lan­guages (e.g. C++) is an in­creas­ing chal­lenge. I'd to re­think how they are built, in­clud­ing the un­der­ly­ing pro­gram­ming lan­guages. In this area, I'm look­ing for can­did­ates who have a sol­id back­ground in the area of pro­gram­ming lan­guages and com­piler con­struc­tion.

Ma­ter­i­al ac­quis­i­tion and mod­el­ing

Ma­ter­i­als found in the real world are beau­ti­ful and of­ten can­not be de­scribed well by simple mod­els used in com­puter graph­ics. I am in­ter­ested in meas­ur­ing ma­ter­i­als to un­der­stand their prop­er­ties, and to use those in­sights to im­prove the mod­els that are cur­rently used. My group has ac­cess to one of the world's most ad­vanced mo­tor­ized go­nio-pho­to­met­ers for ac­quir­ing their op­tic­al prop­er­ties, which is a unique op­por­tun­ity for work in this area. Here are some ex­amples of re­cent work in this area [4] [5].

What EPFL and RGL, and I can of­fer

EPFL provides a world-class, highly col­lab­or­at­ive in­ter­na­tion­al re­search en­vir­on­ment with com­pet­it­ive salar­ies and a beau­ti­ful set­ting on the shores of Lake Geneva. RGL is un­usu­al in that in ad­di­tion to con­duct­ing re­search on fun­da­ment­al as­pects of ren­der­ing, we also in­vest sig­ni­fic­ant amounts of time to build prac­tic­al sys­tems (e.g. Mit­suba, PBRT) that con­sti­tute a widely used found­a­tion of re­search pro­jects around the world. Due to the small size of RGL (cur­rently just 3 PhD can­did­ates), I am able to provide ex­tens­ive su­per­vi­sion to stu­dents–I en­joy re­serving a large por­tion of my time to re­search and soft­ware de­vel­op­ment and of­ten par­ti­cip­ate dir­ectly on some part of their pro­jects.

Doc­tor­al pro­gram

RGL is part of EPFL's School of Com­puter and Com­mu­nic­a­tion Sci­ences (“IC”). Doc­tor­al stu­dents in this school are part of a cent­ral­ized doc­tor­al pro­gram (EDIC), which is some­what un­usu­al com­pared to our European peer in­sti­tu­tions (that said, such pro­grams are com­mon in top-10 schools in the United States).

One im­plic­a­tion of this dif­fer­ence is that stu­dents do not dir­ectly ap­ply to pro­fess­ors (e.g. me), but to a pro­gram. Ad­mis­sion is done by a com­mit­tee of pro­fess­ors from dif­fer­ent areas of com­puter sci­ences with the goal of re­cruit­ing the best can­did­ates. The stu­dents se­lec­ted by this com­mit­tee re­ceive a flex­ible fel­low­ship that cov­ers their first year. Dur­ing this time, they can take classes and con­duct small re­search pro­jects (“EDIC pro­jects” / “ro­ta­tions”) in dif­fer­ent labs, with the goal of choos­ing an ad­visor that will form­ally hire them in­to their lab at the end of the first year. This sys­tem is par­tic­u­larly valu­able if you are in­ter­ested in mul­tiple re­search areas and would like to ex­per­i­ment and gain fur­ther ex­per­i­ence be­fore com­mit­ting to a con­crete path. Mul­tiple ro­ta­tions are not a re­quire­ment, however (in case you are only in­ter­ested in a single lab).

The dead­line of ap­plic­a­tions to EDIC is Decem­ber 15 and con­sists of

The re­com­mend­a­tion let­ters are prob­ably the most im­port­ant part of your ap­plic­a­tion—ideally, they should be from people with whom you have col­lab­or­ated closely, and who can com­ment on your abil­ity to do re­search. Re­com­mend­a­tion let­ters from in­struct­ors who can com­ment on your grades but don't really know you well provide very little in­form­a­tion re­l­at­ive to the grade tran­script. For this reas­on, it's im­port­ant to have at least 1-2 let­ters of the former cat­egory (ideally re­spec­ted col­leagues who can provide a de­tailed and pos­it­ive as­sesse­ment of your qual­it­ies as a re­search­er.)

Please reach out to me if you are in­ter­ested in ap­ply­ing—I'd love to sched­ule a Skype/Zoom call to learn more about your back­ground and in­terests be­fore the EDIC dead­line. The EDIC ap­plic­a­tion form also provides a field where you can spe­cify fac­ulty of in­terest, with whom you have already com­mu­nic­ated—please make sure to enter my name here.