/torch-saliency

code and tools around integral images
  1. Lua 50.7%
  2. C 49.3%
Lua C
Find file
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
generic convert for new luaT specs Sep 19, 2012
CMakeLists.txt remove openmp from CMake as it is automatic now. Sep 5, 2012
README.md more info to README.md Jul 6, 2012
init.lua rename consistent with old name Sep 12, 2012
saliency-1.0-0.rockspec
saliency.c
test.lua convert for new luaT specs Sep 19, 2012

README.md

torch-saliency

code and tools around integral images.

A library for finding interest points based on fast integral historgrams. The saliency is computed based on Kadir 2004. The largest change in entropy between scales is the most salient.

several functions can be of more use generally. The interface will be cleaned up shortly. Need to add a lua interface and help to the working C functions.

compute an integer image ii from l l.libsaliency.intImage(ii,l)

compute an integer histogram rr from l l.libsaliency.intHist(rr,l,nbins,lmin,lmax)

average windows < kr x kc > steped < sr x sc > across the integral image l.libsaliency.intAvg(aa,ii,kr,kc,sr,sc)

Max, OneOverMax, MeanOverMax

rr.libsaliency.spatialMax(mm,rr) rr.libsaliency.spatialOneOverMax(mm,rr) rr.libsaliency.spatialMeanOverMax(mm,rr)

Non Maximal suppression code:

local m,mi,p = saliency.getMax(img,msize,ng)

There is an extensive regression test:

test.lua