view org/adelson-notes.org @ 140:208424fee6f9

new ideas.
author Robert McIntyre <rlm@mit.edu>
date Thu, 05 Feb 2015 23:05:45 -0800
parents dfcbbb3d4b9a
children
line wrap: on
line source
1 #+title: Notes for "Special Topics in Computer Vision"
2 #+author: Robert McIntyre
3 #+email: rlm@mit.edu
4 #+description:
5 #+keywords:
6 #+SETUPFILE: ../../aurellem/org/setup.org
7 #+INCLUDE: ../../aurellem/org/level-0.org
8 #+babel: :mkdirp yes :noweb yes :exports both
10 * Fri Sep 27 2013
12 Lambertian surfaces are a special type of Matt surface. They reflect
13 light in all directions equally. They have only one parameter, the
14 amount of energy that is absorbed/re-emitted.
16 [[../images/adelson-checkerboard.jpg]]
17 #+caption: Lol checkerboard illusion.
19 Look into Helmholtz' stuff, it might be interesting. It was the
20 foundation of both vision and audition research. Seems to have took
21 a sort of Baysean approach to inferring how vision/audition works.
23 - Homomorphic filtering :: Oppenhiem, Schafer, Stockham, 1968. also
24 look at Stockham, 1972.
26 Edwin Land was Adelson's hero back in the day. He needed to create a
27 color photo for the Polaroid camera. In order to process for
28 automatic development of film, he had to get a good approximation for
29 the illumination/reflectance decomposition that humans do, which he
30 called Retinex.
32 Cornsweet square wave grating is cool.
34 - Retinex :: use derivatives to find illumination. Sort of
35 implicitly deals with edges, etc. Can't deal with
36 non-lambertian objects.
39 Adelson introduces the problem as an "inverse" problem, where you
40 try to "undo" the 3-d projection of the world on your retina.
42 On the functional view of vision : "What it takes" is to build a
43 model of the world in your head. The bare minimum to get success in
44 life is to have a model of the world. Even at the level of a single
45 cell, I think you still benefit from models.
47 Spatial propagation is ABSOLUTELY required to separate embossed
48 stuff from "painted" stuff. Edges, likewise, MUST have spatial
49 context to disambiguate. The filters we use to deal with edges must
50 have larger spatial context to work, and the spatial extent of this
51 context must be the ENTIRE visual field in some cases!
53 ------------------------------------------------------------
55 ** Illumination, shape, reflectance all at once
57 What if we tried to infer everything together? Some images are so
58 ambiguous it requires propagation from all three qualities to
59 resolve the ambiguity.
61 Brain has a competing painter, sculptor, and gaffer which each try
62 to "build" the things in the world. There is a cost to everything
63 such as paints, lights, and material, and then you try to optmize
64 some cost function using these primitives.
67 Horn, technical report, 1970
69 * Fri Oct 4 2013
71 Student report. Talked about how you capture the appearance of a
72 grape. It's actually quite compicated, involving gloss, spatial
73 context, etc.
75 Turbosquid seems interesting. They sell 3D models of stuff.
77 BRDF -- bi-directional reflectance distribution function this shows
78 how a surface will behave given lighting conditions. Lambertian is a
79 simple parameterized instantiation of this.
81 BSSRDF -- (SS = subsurface) 3D analogue of BRDF
83 What would the 3D analiogue of texture be?
85 (a : b : c) as (a + b + c : b + c : c) <-- this is just the golden
86 ratio again!
88 CURET BTF Database lol what's this
90 This student went and gathered 1000 images of different large
91 objects made of different materials. The images were gathered off of
92 Flikr.
94 Then she gave another talk from someone else. It's about assigning
95 materials to objects and then rendering them. The choice of
96 materials is determined by some sort of expert system?
98 They have made a neat looking interface for human entry of texture
99 labeling or objects in scenes. The important elements were manual
100 labels, dynamic display of the current selection, and undo.
102 There are papers about mechanical turk engineering.
104 lol http://opensurfaces.cs.cornell.edu/
106 CUBAM is interesting 2010 "The multidimensional wisdom of crowds."
107 "Neural Information processing systems" some sort of voting scheme.
109 The point of this is apparently to do some kitchen makeover
110 thing. You would take a picture of your kitchen, or you would look
111 for a kitchen that looks like yours, and then you would be able to
112 investigate different textures for your own kitchen.
114 apparently "Label ME" has never been appropriately
115 crowdsourced. Turns out that you get better work if you don't use
116 idiots to do the work lol.