BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Robust Shape Acquisition with Differentiable Rendering
DTSTART:20210712T140000
DTEND:20210712T160000
DTSTAMP:20260502T223121Z
UID:4ad38c1b5cbb9940851ca57758412f526b5b56015e049c05fc94d148
CATEGORIES:Conferences - Seminars
DESCRIPTION:Baptiste Nicolet\nEDIC candidacy exam\nexam president: Prof. M
 ark Pauly\nthesis advisor: Prof. Wenzel Jakob\nco-examiner: Prof. Amir Zam
 ir\n\nAbstract\nDifferentiable rendering aims at computing derivatives of 
 userspecified\nobjectives with respect to scene parameters\, by differenti
 ating\nthe light transport simulation\, thus allowing to solve challenging
 \ninverse rendering problems. In recent years\, considerable progress\nhas
  been made towards that goal\, allowing to design correct estimators\nof d
 erivative quantities.\nHowever\, running optimizations with differentiable
  rendering\nremains challenging\, especially when it comes to optimizing g
 eometry.\nSpecial care must be taken to avoid bias in the estimated\ngradi
 ents\, and the high sparsity of the gradients obtained make geometry\nopti
 mizations tedious. In this report we review three recent\npapers that tack
 le the issue of geometric discontinuities\, and review\nour work on counte
 racting the sparsity of the resulting gradients.\n\nBackground papers\n\n	
 Path Space Differentiable Rendering : link\n	Unbiased Warped Area Samplin
 g for Differentiable Rendering : link (errata)\n	Monte Carlo Estimators f
 or Differential Light Transport : link\n
LOCATION:
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR