1.1 --- /dev/null	Thu Jan 01 00:00:00 1970 +0000
     1.2 +++ b/org/adelson-notes.org	Sun Oct 13 12:09:05 2013 -0400
     1.3 @@ -0,0 +1,98 @@
     1.4 +#+title: Notes for "Special Topics in Computer Vision"
     1.5 +#+author: Robert McIntyre
     1.6 +#+email: rlm@mit.edu
     1.7 +#+description: 
     1.8 +#+keywords: 
     1.9 +#+SETUPFILE: ../../aurellem/org/setup.org
    1.10 +#+INCLUDE: ../../aurellem/org/level-0.org
    1.11 +#+babel: :mkdirp yes :noweb yes :exports both
    1.12 +
    1.13 +* Fri Sep 27 2013
    1.14 +
    1.15 +  Lambertian surfaces are a special type of Matt surface. They reflect
    1.16 +  light in all directions equally. They have only one parameter, the
    1.17 +  amount of energy that is absorbed/re-emitted.
    1.18 +
    1.19 +  [[../images/adelson-checkerboard.jpg]]
    1.20 +  #+caption: Lol checkerboard illusion.
    1.21 +
    1.22 +  Look into Helmholtz' stuff, it might be interesting. It was the
    1.23 +  foundation of both vision and audition research. Seems to have took
    1.24 +  a sort of Baysean approach to inferring how vision/audition works.
    1.25 +  
    1.26 +  - Homomorphic filtering :: Oppenhiem, Schafer, Stockham, 1968.  also
    1.27 +       look at Stockham, 1972.
    1.28 +
    1.29 +  Edwin Land was Adelson's hero back in the day. He needed to create a
    1.30 +  color photo for the Polaroid camera. In order to process for
    1.31 +  automatic development of film, he had to get a good approximation for
    1.32 +  the illumination/reflectance decomposition that humans do, which he
    1.33 +  called Retinex.
    1.34 +
    1.35 +  Cornsweet square wave grating is cool.
    1.36 +
    1.37 +  - Retinex :: use derivatives to find illumination. Sort of
    1.38 +               implicitly deals with edges, etc. Can't deal with
    1.39 +               non-lambertian objects. 
    1.40 +
    1.41 +
    1.42 +  Adelson introduces the problem as an "inverse" problem, where you
    1.43 +  try to "undo" the 3-d projection of the world on your retina.
    1.44 +
    1.45 +  On the functional view of vision : "What it takes" is to build a
    1.46 +  model of the world in your head. The bare minimum to get success in
    1.47 +  life is to have a model of the world. Even at the level of a single
    1.48 +  cell, I think you still benefit from models.
    1.49 +
    1.50 +  Spatial propagation is ABSOLUTELY required to separate embossed
    1.51 +  stuff from "painted" stuff. Edges, likewise, MUST have spatial
    1.52 +  context to disambiguate. The filters we use to deal with edges must
    1.53 +  have larger spatial context to work, and the spatial extent of this
    1.54 +  context must be the ENTIRE visual field in some cases!
    1.55 +
    1.56 +------------------------------------------------------------
    1.57 +
    1.58 +** Illumination, shape, reflectance all at once
    1.59 +   
    1.60 +   What if we tried to infer everything together? Some images are so
    1.61 +   ambiguous it requires propagation from all three qualities to
    1.62 +   resolve the ambiguity.
    1.63 +
    1.64 +   Brain has a competing painter, sculptor, and gaffer which each try
    1.65 +   to "build" the things in the world. There is a cost to everything
    1.66 +   such as paints, lights, and material, and then you try to optmize
    1.67 +   some cost function using these primitives.
    1.68 +
    1.69 +
    1.70 +   Horn, technical report, 1970
    1.71 +
    1.72 +
    1.73 +
    1.74 +* Fri Oct  4 2013
    1.75 +
    1.76 +  Student report. Talked about how you capture the appearance of a
    1.77 +  grape. It's actually quite compicated, involving gloss, spatial
    1.78 +  context, etc.
    1.79 +
    1.80 +  Turbosquid seems interesting. They sell 3D models of stuff.
    1.81 +
    1.82 +  BRDF -- bi-directional reflectance distribution function this shows
    1.83 +  how a surface will behave given lighting conditions. Lambertian is a
    1.84 +  simple parameterized instantiation of this.
    1.85 +
    1.86 +  BSSRDF -- (SS = subsurface) 3D analogue of BRDF
    1.87 +
    1.88 +  What would the 3D analiogue of texture be?
    1.89 +
    1.90 +  (a : b : c) as (a + b + c : b + c : c) <-- this is just the golden
    1.91 +  ratio again!
    1.92 +
    1.93 +  CURET BTF Database lol what's this
    1.94 +
    1.95 +  This student went and gathered 1000 images of different large
    1.96 +  objects made of different materials. The images were gathered off of
    1.97 +  Flikr. 
    1.98 +
    1.99 +  
   1.100 +
   1.101 +