This item is only available for download by members of the University of Illinois community. Students, faculty, and staff at the U of I may log in with your NetID and password to view the item. If you are trying to access an Illinois-restricted dissertation or thesis, you can request a copy through your library's Inter-Library Loan office or purchase a copy directly from ProQuest.
Permalink
https://hdl.handle.net/2142/107273
Description
Title
Simple image depth and BRDF estimation
Author(s)
Jeong, Dohun
Contributor(s)
Do, Minh
Issue Date
2020-05
Keyword(s)
Differentiable Rendering
Variational Inference
Computer Vision
Depth Estimation
BRDF
Abstract
Classical computer vision algorithms for scene reconstructions have restrictive assumptions about
the scene, such as the absence of specular reflection and indirect lighting. However, in most
real world photographs, these assumptions often fail. Recent advancement in physically based
rendering and neural networks is changing how we can infer scene parameters from a single
photograph. Solving the rendering equation to simulate the light transport allows us to synthesize
photorealistic images from physical scene parameters. These images take indirect lighting,
specular reflections, and other phenomena into account. Differentiable rendering incorporates
auto-differentiation and backpropagation to optimize the physical scene parameters based
on the derivatives of the rendering equation with respect to the scene parameters. Free from
the restrictions of classical computer vision algorithms, differentiable rendering can be used to
reconstruct a 3D scene, and perform relighting, material editing, and other applications. This
thesis investigates the recovery of intrinsic scene parameters through differentiable rendering and
variational inference. The bi-directional reflectance function (BRDF) of materials are estimated using
a flash-no flash image pair and scene geometry input. With the ubiquity of depth cameras and time
of flight sensors in new hardware, as well as 3D indoor geometry reconstruction algorithms, this thesis assumes known scene geometry. Depth is estimated through variational inference of the
collapsed dimension, by finding a hidden pattern in the latent space. This closely follows a recent
study on visual deprojection to recover a collapsed dimension of an image. Combined, estimated
BRDF and depth can decompose images in Intrinsic Image Decomposition, and derive normal
maps, albedo estimation, and beyond.
Use this login method if you
don't
have an
@illinois.edu
email address.
(Oops, I do have one)
IDEALS migrated to a new platform on June 23, 2022. If you created
your account prior to this date, you will have to reset your password
using the forgot-password link below.