Contour integration classes and programs.
More...
Files | |
| file | cascadeHold.H [code] |
| file | CINNIC.C [code] |
| file | CINNIC.H [code] |
| file | CINNIC2.C [code] |
| file | CINNIC2.H [code] |
| file | CINNICanova.C [code] |
| file | CINNICinteract.C [code] |
| file | CINNICstats.C [code] |
| file | CINNICstatsRun.C [code] |
| file | CINNICstatsRun.H [code] |
| file | CINNICtest.C [code] |
| file | contourDefine.H [code] |
| file | contourNeuron.C [code] |
| file | contourNeuron.H [code] |
| file | contourNeuronProp.C [code] |
| file | contourNeuronProp.H [code] |
| file | contourNeuronProp2.H [code] |
| file | contourRun.C [code] |
| file | contourRun.H [code] |
| file | contourRun2.C [code] |
| file | contourRun2.H [code] |
| file | gaborElement.C [code] |
| file | gaborElement.H [code] |
| file | ImageMap.H [code] |
| file | MrGabor.C [code] |
| file | MrGabor2.C [code] |
| file | pointAndFlood.C [code] |
| file | PolatSagiTest.C [code] |
| file | README.dxy [code] |
| file | runCINNIC2.C [code] |
| file | staticContourNeuronProp.H [code] |
| file | statsIncl.H [code] |
Contour integration classes and programs.
This is an over view and documentation of the CINNIC classes and files.
CINNICtest.C - This is the standard entry binary exec for CINNIC. it makes calles to create the 4D kernal and then can display the kernal properties if no command line is given or it can call the full CINNIC alg. with proper command line args.
example: CINNICtest input.pgm output_files configfile where input.pgm is a PGM file, ouput_files is a prefix for a file and its directory for all output files and configFile is a valid config file for CINNIC (see) contour.conf
cascadeHold.H - Holder for cascade energys. Is most likely depricated
readConfig.H/.C - A generic class for opening and parsing config files
CINNIC.H/.C - These are entry level methods into CINNIC. For instance you can use this to view the neuron template (Kernel). However, you will more than likely call RunSimpleIMage which is a call to the main CINNIC loop. The image is rescaled,filtered and in general pre-processed.
PreImage - is responsible for setting group sizes.
runImage - is the inner most loop of this class and calls contourRunMain which does the CINNIC neural implementations and processes results and writes most (but not all) of the results to disk. Results are writen as both image (e.g ss.max.out.d.d.d.d) and as a tab-delim file with more precise values (ss.table.out.txt) the later is used for precise analysis for instance Polat and Sagi 1993
contourRun.H/.C - This is the meat of the CINNIC package. This is where the neurons operate on the orientation filtered images. Group values are computered here as will as the fast plasticity term. Much old code still exists for doing cascade in CINNIC that is no longer used. It will not be discused. The flow here is dedicated to iterating over the kernals and orientation filtered images. The iterations are handled outside on CINNIC.H/.C (i.e. methods this class will be called once for each iteration). Many small methods exist here for handeling the saliency map as well as sigmoidal functions and the various storage matricses used in testing.
contourRunMain - Mainly resets verious data structures at the first iteration then calls runImageSigmoid
runImageSigmoid - This is the out loop for the 6 layer psuedo-convolution. It also controls the group supression modifier. Here the change in a groups potential is detected from one iteration to the next. This adjusts the group supression. It next calls iterateConvolve which manages the inner loop for the CINNIC pseudo convolution.
iterateConvolve - Simply decides if cascade should be used calls convolveSimple which is the inner loop
convolveSimple - Is the inner loop. It manages the fast plasticity term for this neuron. This loop is i,j,k,l while the outer loop is alpha,beta. Optimizations and tests are used here to skip near zero values. THE TRICK here is that the result is stored in the other neuron (I had my reasons for doing it this way) that is each simple interaction is stored in the neuron that is in defined in the outer loop. This means that the result is put into beta,k,l while alpha,i,j is treated on the fly. beta,k,l and alpha,i,j are reversed in the paper from what they are called here.
All potential is stored in the NeuronMatrix which has the form matrix[iteration][beta][k][l]. This is of type ContourNeuronProp.
contourNeuronProp.H - This is a basic class for holding the propertie of each neuron. In essence it holds the potential of each neuron. It was initially designed so that cascade values could be tracked. Since it is imbedded in the code it has stayed for a little while. It should be replaced at some point by a more efficent data structure. Important methods are getCharge and Charge.
contourNeuron.H/.C - This creates and then stores the generic neuron template kernal in PropHold. This has a variety of crazy experimental calculations that are used in the creation of the CINNIC kernel. note that values of PropHold are stored in the 4D neural matrix called FourDNeuralMap[i][j][k][l] where i and j are the alpha/beta angles and k and l are the coords in each kernel slice.
-----
Other semi-related classes:
CINNICstats - is a basic statistical anal. package.
CINNICstatsRun - executes some states used by CINNICstats
CINNICanova - an analysis of variance package
for dependency graphs: rankdir: TB
1.6.3