Modify aggregate channel features object detector. Takes an object detector trained by acfTrain() and modifies it. Only certain modifications are allowed to the detector and the detector should never be modified directly (this may cause the detector to be invalid and cause segmentation faults). Any valid modification to a detector after it is trained should be performed using acfModify(). The parameters 'nPerOct', 'nOctUp', 'nApprox', 'lambdas', 'pad', 'minDs' modify the channel feature pyramid created (see help of chnsPyramid.m for more details) and primarily control the scales used. The parameters 'pNms', 'stride', 'cascThr' and 'cascCal' modify the detector behavior (see help of acfTrain.m for more details). Finally, 'rescale' can be used to rescale the trained detector (this change is irreversible). USAGE detector = acfModify( detector, pModify ) INPUTS detector - detector trained via acfTrain pModify - parameters (struct or name/value pairs) .nPerOct - [] number of scales per octave .nOctUp - [] number of upsampled octaves to compute .nApprox - [] number of approx. scales to use .lambdas - [] coefficients for power law scaling (see BMVC10) .pad - [] amount to pad channels (along T/B and L/R) .minDs - [] minimum image size for channel computation .pNms - [] params for non-maximal suppression (see bbNms.m) .stride - [] spatial stride between detection windows .cascThr - [] constant cascade threshold (affects speed/accuracy) .cascCal - [] cascade calibration (affects speed/accuracy) .rescale - [] rescale entire detector by given ratio OUTPUTS detector - modified object detector EXAMPLE See also chnsPyramid, bbNms, acfTrain, acfDetect Piotr's Computer Vision Matlab Toolbox Version 3.20 Copyright 2014 Piotr Dollar. [pdollar-at-gmail.com] Licensed under the Simplified BSD License [see external/bsd.txt]