vcm
Neural
The call script is :-
vcm.Neural(clip,[str txt = "none"[ ,str fname][,clip tclip]] [,int iter = 200][ ,int bestof = 1][, int wset = 0][,int xpts = 3][, int ypts = xpts][, int tlx = xpts][, int trx = width - xpts][, int tty = ypts][, int tby = height - ypts][, int rgb = 1])
clip: Inpt clip all formats except Compat accepted.
txt: can be either "none", "read" or "save". In case of read or save fname with full path and file name be given to read from or save to. If "none" it neither saves nor reads a file.
if "save" relevant parameters and training progress (error sum) will be saved to a file. If "read", then it gets all parameters and best weights to use from the previously saved file. None of following parameters need to be specified then.
tclip: clip on which desired data is available. Neural finds optimum weights to match the tclip.
iter: number of iterations for training
bestof: best of is number of times process repeated with new starting weight sets and use wights which resulted in least error.
wset: if set to 0, uses internal random numbers and If set to 1, uses current time seeded random numbers as starting wights.
xpts, ypts: number of input points along width and height. A max of xpts * ypts = 225 and minimum of 9 can be specified. xpts and ypts must be odd positive numbers.
tlx, trx, tty, tby: top left x, right x top y and bottom y coordinates of window for training. Time for training is proportional to xpts * ypts * (trx - tlx - xpts) * (tby - tty - ypts) * iter * bestof.
rgb: with RGB input if set to 0, red, if set to 1 green and if set to 2 blue color plane is used for training. For YUV, or Gray only Y is used. processed.
Given an input ( first frame of clip) and its corresponding desired output ( first frame of tclip which has been obtained through some linear processing),
function Neural trains itself to get an output as close to desired output as possible. Training window should represent all possible input variations and be as small as possible. Given input grid size ( xpts * ypts) to work on and corresonding desred center points, through repeated iterations and different starting weight sets it finds weights that result in minimum error. The weight set can be saved in a text file and later used for same or similar data . This saves training time that can take several seconds to a few minutes. The input clip is then processed using these weights.