Scripts for VLBenchmark evaluation checked in

vlbenchmark/KAZE.m

@@ -0,0 +1,120 @@
+% KAZE.m
+% Class for testing KAZE with the vlbenchmark
+% Author: Pablo F. Alcantarilla
+% Date: 27-12-2015
+% Note: The code is an adaptation of the original code from Karel Lenc
+%**************************************************************************
+classdef KAZE < localFeatures.GenericLocalFeatureExtractor & ...
+    helpers.GenericInstaller
+
+properties (SetAccess=public, GetAccess=public)
+    Opts = struct(...
+      'soffset', 1.30,...        % the base scale offset (sigma units)
+      'omax', 4,...              % the coarsest nonlinear scale space level (sigma units)
+      'nsublevels', 4,...        % number of sublevels per octave
+      'dthreshold', .0008,...    % Feature detector threshold response for accepting points
+      'descriptor', 5,...        % Descriptor Type (See KAZEConfig.h for more info)
+      'sderivatives', 1,...      % Gaussian smoothing for the derivatives     
+      'diffusivity', 1,...       % Diffusivity type: 0->PM_G1, 1->PM_G2, 2->Weickert, 3->Charbonnier
+      'show_results', 1,...      % 1 in case we want to visualize the detection results
+      'save_scale_space', 0 ...  % 1 in case we want to save the nonlinear scale space images. 0 otherwise
+      );
+  end
+
+  properties (Constant,Hidden)
+    RootInstallDir = fullfile('data','software','kaze','');
+    SrcDir = fullfile(localFeatures.Kaze.RootInstallDir,'');
+    BinPath = fullfile(localFeatures.Kaze.SrcDir,'kaze_features');
+  end
+
+  methods
+    function obj = KAZE(varargin)
+      import helpers.*;
+      obj.Name = 'KAZE';
+      obj.ExtractsDescriptors = true;
+      varargin = obj.checkInstall(varargin);
+      varargin = obj.configureLogger(obj.Name,varargin);
+      obj.Opts = vl_argparse(obj.Opts, varargin);
+    end
+
+    function [frames descriptors] = extractFeatures(obj, origImagePath)
+      import helpers.*;
+      import localFeatures.helpers.*;
+      frames = obj.loadFeatures(origImagePath,true);
+      if numel(frames) > 0; return; end;      
+      obj.info('Computing features of image %s.',...
+        getFileName(origImagePath));
+      [imagePath imIsTmp] = obj.ensureImageFormat(origImagePath);
+      tmpName = tempname;
+      outFeaturesFile = [tmpName '.kaze'];
+      args = obj.buildArgs(imagePath, outFeaturesFile);
+      cmd = [obj.BinPath ' ' args];
+      obj.debug('Executing: %s',cmd);
+      startTime = tic;
+      [status,msg] = helpers.osExec('.',cmd,'-echo');
+      if status ~= 0
+        error('%d: %s: %s', status, cmd, msg) ;
+      end
+      timeElapsed = toc(startTime);
+      [frames descriptors] = ...
+        readFeaturesFile(outFeaturesFile,'FloatDesc',true);
+      delete(outFeaturesFile);
+      if imIsTmp, delete(imagePath); end;
+      obj.debug('%d features from image %s computed in %gs',...
+        size(frames,2),getFileName(imagePath),timeElapsed);
+      obj.storeFeatures(imagePath, frames, []);
+    end
+
+    function sign = getSignature(obj)
+      sign = [helpers.fileSignature(obj.BinPath) ';'...
+              helpers.struct2str(obj.Opts)];
+    end
+  end
+
+  methods (Access=protected)
+    function args = buildArgs(obj, imagePath, outFile)
+      % -nl - do not write the hessian keypoint type
+      args = sprintf(' "%s" --output "%s"',...
+        imagePath, outFile);
+      fields = fieldnames(obj.Opts);
+      for i = 1:numel(fields)
+        val = obj.Opts.(fields{i});
+        if ~isempty(val)
+          args = [args,' --',fields{i},' ', num2str(val)];
+        end
+      end
+    end
+
+    function deps = getDependencies(obj)
+      deps = {helpers.Installer()};
+    end
+
+    function res = isCompiled(obj)
+      res = exist(obj.BinPath,'file');
+    end
+
+    function compile(obj)
+      % Run cmake; make and make install
+      import helpers.*;
+      if obj.isCompiled()
+        return;
+      end
+      if ~exist(obj.SrcDir,'dir')
+        error('Source code of KAZE feature detector not present in %s.',...
+          obj.SrcDir);
+      end
+
+      fprintf('Compiling KAZE\n');
+      % Run cmake with sys. libraries environment
+      [status msg] = helpers.osExec(obj.SrcDir,'cmake .','-echo');
+      if status ~= 0
+        error('CMake error: \n%s',msg);
+      end
+      % Run Make
+      [status msg] = helpers.osExec(obj.SrcDir,'make','-echo');
+      if status ~= 0
+        error('Make error: \n%s',msg);
+      end
+    end
+  end
+end

vlbenchmark/README

@@ -0,0 +1,23 @@
+The code in this folder is for KAZE evaluation in the VLBenchmarks.
+I assume you have Matlab installed in your computer. The next steps are:
+
+1. Download VLBenchmarks in your computer:
+http://www.vlfeat.org/benchmarks/
+
+2. Compile akaze_features and move the executable file to: 
+/vlfeat-vlbenchmarks/data/software/kaze/
+
+3. Copy the file KAZE.m to:
+/vlfeat-vlbenchmarks/+localFeatures/
+
+4. Copy the file RepeatabilityBenchmarkMod.m to:
+/vlfeat-vlbenchmarks/+benchmarks/
+
+RepeatabilityBenchmarkMod contains some modifications with respect to the original 
+RepeatabilityBenchmark, such as checking pixel location error.
+
+5. Copy the script test_detector_bikes_akaze.m to:
+/vlfeat-vlbenchmarks/
+
+Run the script to evaluate KAZE in the bikes dataset. You can change the script for 
+evaluation in other datasets.