1.1 --- a/org/gabor.org	Tue Mar 12 04:34:37 2013 +0000
     1.2 +++ b/org/gabor.org	Thu Apr 11 05:40:23 2013 +0000
     1.3 @@ -164,10 +164,7 @@
     1.4  (draw-kernel! (gabor-kernel 50 4 (/ Math/PI 3) 3 0)
     1.5                (str img-base "gabor-50-4-pi-over3-3.png"))
     1.6  #+end_src
     1.7 -
     1.8 -
     1.9 -
    1.10 -     
    1.11 +   
    1.12  
    1.13  #+name: gabor-tail
    1.14  #+begin_src clojure

     2.1 --- a/org/ideas.org	Tue Mar 12 04:34:37 2013 +0000
     2.2 +++ b/org/ideas.org	Thu Apr 11 05:40:23 2013 +0000
     2.3 @@ -116,3 +116,11 @@
     2.4  ;;Builders wrought with greatest care
     2.5  ;;Each minute and unseen part;
     2.6  ;;For the Gods see everywhere.
     2.7 +
     2.8 +
     2.9 +* misc
    2.10 +  - use object tracking on moving objects to derive good static
    2.11 +    detectors and achieve background separation
    2.12 +  - temporal scale pyramids.  this can help in verb recognition by
    2.13 +    making verb identification time-scale independent (up to a certian
    2.14 +    factor)
    2.15 \ No newline at end of file

     3.1 --- a/org/literature-review.org	Tue Mar 12 04:34:37 2013 +0000
     3.2 +++ b/org/literature-review.org	Thu Apr 11 05:40:23 2013 +0000
     3.3 @@ -42,13 +42,13 @@
     3.4  
     3.5   - Zhang, Z., R. Deriche, O. Faugeras, Q.T. Luong, “A robust
     3.6    technique for matching two uncalibrated images through the recovery
     3.7 -  of the unknown epipolar geometry,” Artificial In- telligence, 78,
     3.8 +  of the unknown epipolar geometry,” Artificial Intelligence, 78,
     3.9    (1995), pp. 87-119.
    3.10  
    3.11  
    3.12  
    3.13  
    3.14 -
    3.15 +   
    3.16  * Alignment by Maximization of Mutual Information, Paul A. Viola
    3.17  
    3.18    PhD Thesis recommended by Winston. Describes a system that is able
    3.19 @@ -63,18 +63,198 @@
    3.20    - Differential entropy seems a bit odd -- you would think that it
    3.21      should be the same as normal entropy for a discrete distrubition
    3.22      embedded in continuous space. How do you measure the entropy of a
    3.23 -    half continuous, half discrete random variable?
    3.24 +    half continuous, half discrete random variable? Perhaps the
    3.25 +    problem is related to the delta function, and not the definition
    3.26 +    of differential entropy?
    3.27  
    3.28    - Expectation Maximation (Mixture of Gaussians cool stuff)
    3.29      (Dempster 1977)
    3.30  
    3.31    - Good introduction to Parzen Window Density Estimation. Parzen
    3.32      density functions trade construction time for evaulation
    3.33 -    time.(Pg. 41)
    3.34 +    time.(Pg. 41) They are a way to transform a sample into a
    3.35 +    distribution. They don't work very well in higher dimensions due
    3.36 +    to the thinning of sample points.
    3.37 +
    3.38 +  - Calculating the entropy of a Markov Model (or state machine,
    3.39 +    program, etc) seems like it would be very hard, since each trial
    3.40 +    would not be independent of the other trials. Yet, there are many
    3.41 +    common sense models that do need to have state to accurately model
    3.42 +    the world.
    3.43 +
    3.44 +  - "... there is no direct procedure for evaluating entropy from a
    3.45 +    sample. A common approach is to model the density from the sample,
    3.46 +    and then estimate the entropy from the density."
    3.47 +
    3.48 +  - pg. 55 he says that infinity minus infinity is zero lol.
    3.49 +
    3.50 +  - great idea on pg 62 about using random samples from images to
    3.51 +    speed up computation.
    3.52 +
    3.53 +  - practical way of terminating a random search: "A better idea is to
    3.54 +    reduce the learning rate until the parameters have a reasonable
    3.55 +    variance and then take the average parameters."
    3.56 +
    3.57 +  - p. 65 bullshit hack to make his parzen window estimates work.
    3.58 +
    3.59 +  - this alignment only works if the initial pose is not very far
    3.60 +    off. 
    3.61 +
    3.62  
    3.63    Occlusion? Seems a bit holistic.
    3.64  
    3.65 +** References
    3.66 + - "excellent" book on entropy (Cover & Thomas, 1991) [Elements of
    3.67 +   Information Theory.] 
    3.68 +
    3.69 + - Canny, J. (1986). A Computational Approach to Edge Detection. IEEE
    3.70 +   Transactions PAMI, PAMI-8(6):679{698
    3.71 +
    3.72 + - Chin, R. and Dyer, C. (1986). Model-Based Recognition in Robot
    3.73 +   Vision. Computing Surveys, 18:67-108.
    3.74 +
    3.75 + - Grimson, W., Lozano-Perez, T., Wells, W., et al. (1994). An
    3.76 +   Automatic Registration Method for Frameless Stereotaxy, Image
    3.77 +   Guided Surgery, and Enhanced Realigy Visualization. In Proceedings
    3.78 +   of the Computer Society Conference on Computer Vision and Pattern
    3.79 +   Recognition, Seattle, WA. IEEE.
    3.80 +
    3.81 + - Hill, D. L., Studholme, C., and Hawkes, D. J. (1994). Voxel
    3.82 +   Similarity Measures for Auto-mated Image Registration. In
    3.83 +   Proceedings of the Third Conference on Visualization in Biomedical
    3.84 +   Computing, pages 205 { 216. SPIE.
    3.85 +
    3.86 + - Kirkpatrick, S., Gelatt, C., and Vecch Optimization by Simulated
    3.87 +   Annealing. Science, 220(4598):671-680.
    3.88 +
    3.89 + - Jones, M. and Poggio, T. (1995). Model-based matching of line
    3.90 +   drawings by linear combin-ations of prototypes. Proceedings of the
    3.91 +   International Conference on Computer Vision
    3.92 +
    3.93 + - Ljung, L. and Soderstrom, T. (1983). Theory and Practice of
    3.94 +   Recursive Identi cation. MIT Press.
    3.95 +
    3.96 + - Shannon, C. E. (1948). A mathematical theory of communication. Bell
    3.97 +   Systems Technical Journal, 27:379-423 and 623-656.
    3.98 +
    3.99 + - Shashua, A. (1992). Geometry and Photometry in 3D Visual
   3.100 +   Recognition. PhD thesis, M.I.T Artificial Intelligence Laboratory,
   3.101 +   AI-TR-1401.
   3.102 +
   3.103 + - William H. Press, Brian P. Flannery, S. A. T. and Veterling,
   3.104 +   W. T. (1992). Numerical Recipes in C: The Art of Scienti c
   3.105 +   Computing. Cambridge University Press, Cambridge, England, second
   3.106 +   edition edition.
   3.107 +
   3.108 +* Semi-Automated Dialogue Act Classification for Situated Social Agents in Games, Deb Roy 
   3.109 +  
   3.110 +  Interesting attempt to learn "social scripts" related to resturant
   3.111 +  behaviour. The authors do this by creating a game which implements a
   3.112 +  virtual restruant, and recoding actual human players as they
   3.113 +  interact with the game. The learn scripts from annotated
   3.114 +  interactions and then use those scripts to label other
   3.115 +  interactions. They don't get very good results, but their
   3.116 +  methodology of creating a virtual world and recording
   3.117 +  low-dimensional actions is interesting.
   3.118 +
   3.119 +  - Torque 2D/3D looks like an interesting game engine.
   3.120 +
   3.121 +
   3.122 +* Face Recognition by Humans: Nineteen Results all Computer Vision Researchers should know, Sinha
   3.123 +  
   3.124 +  This is a summary of a lot of bio experiments on human face
   3.125 +  recognition.
   3.126 +  
   3.127 +  - They assert again that the internal gradients/structures of a face
   3.128 +    are more important than the edges.
   3.129 +
   3.130 +  - It's amazing to me that it takes about 10 years after birth for a
   3.131 +    human to get advanced adult-like face detection. They go through
   3.132 +    feature based processing to a holistic based approach during this
   3.133 +    time.
   3.134 +
   3.135 +  - Finally, color is a very important cue for identifying faces.
   3.136  
   3.137  ** References
   3.138 - - "excellent" book on entropy (Cover & Thomas, 1991)
   3.139 - 
   3.140 \ No newline at end of file
   3.141 +  - A. Freire, K. Lee, and L. A. Symons, BThe face-inversion effect as
   3.142 +    a deficit in the encoding of configural information: Direct
   3.143 +    evidence,[ Perception, vol. 29, no. 2, pp. 159–170, 2000.
   3.144 +  - M. B. Lewis, BThatcher’s children: Development and the Thatcher
   3.145 +    illusion,[Perception, vol. 32, pp. 1415–21, 2003.
   3.146 +  - E. McKone and N. Kanwisher, BDoes the human brain process objects
   3.147 +    of expertise like faces? A review of the evidence,[ in From Monkey
   3.148 +    Brain to Human Brain, S. Dehaene, J. R. Duhamel, M. Hauser, and
   3.149 +    G. Rizzolatti, Eds. Cambridge, MA: MIT Press, 2005.
   3.150 +
   3.151 +
   3.152 +
   3.153 +
   3.154 +heee~eeyyyy kids, time to get eagle'd!!!!
   3.155 +
   3.156 +
   3.157 +
   3.158 +
   3.159 +
   3.160 +* Ullman 
   3.161 +
   3.162 +Actual code reuse!
   3.163 +
   3.164 +precision = fraction of retrieved instances that are relevant
   3.165 +  (true-postives/(true-positives+false-positives))
   3.166 +
   3.167 +recall    =  fraction of relevant instances that are retrieved
   3.168 +  (true-positives/total-in-class)
   3.169 +
   3.170 +cross-validation = train the model on two different sets to prevent
   3.171 +overfitting. 
   3.172 +
   3.173 +
   3.174 +
   3.175 +
   3.176 +
   3.177 +** Getting around the dumb "fixed training set" methods
   3.178 +
   3.179 +*** 2006 Learning to classify by ongoing feature selection
   3.180 +    
   3.181 +    Brings in the most informative features of a class, based on
   3.182 +    mutual information between that feature and all the examples
   3.183 +    encountered so far. To bound the running time, he uses only a
   3.184 +    fixed number of the most recent examples. He uses a replacement
   3.185 +    strategy to tell whether a new feature is better than one of the
   3.186 +    corrent features.
   3.187 +
   3.188 +*** 2009 Learning model complexity in an online environment
   3.189 +    
   3.190 +    Sort of like the heirichal baysean models of Tennanbaum, this
   3.191 +    system makes the model more and more complicated as it gets more
   3.192 +    and more training data. It does this by using two systems in
   3.193 +    parallell and then whenever the more complex one seems to be
   3.194 +    needed by the data, the less complex one is thrown out, and an
   3.195 +    even more complex model is initialized in its place.
   3.196 +
   3.197 +    He uses a SVM with polynominal kernels of varying complexity. He
   3.198 +    gets good perfoemance on a handwriting classfication using a large
   3.199 +    range of training samples, since his model changes complexity
   3.200 +    depending on the number of training samples. The simpler models do
   3.201 +    better with few training points, and the more complex ones do
   3.202 +    better with many training points.
   3.203 +
   3.204 +    The more complex models must be able to be initialized efficiently
   3.205 +    from the less complex models which they replace!
   3.206 +
   3.207 +
   3.208 +** Non Parametric Models
   3.209 +
   3.210 +*** Visual features of intermediate complexity and their use in classification
   3.211 +
   3.212 +*** The chains model for detecting parts by their context
   3.213 +
   3.214 +    Like the constelation method for rigid objects, but extended to
   3.215 +    non-rigid objects as well.
   3.216 +
   3.217 +    Allows you to build a hand detector from a face detector. This is
   3.218 +    usefull because hands might be only a few pixels, and very
   3.219 +    ambiguous in an image, but if you are expecting them at the end of
   3.220 +    an arm, then they become easier to find.
   3.221 +
   3.222 +    
   3.223 \ No newline at end of file