www.gusucode.com > datastoreio工具箱 matlab源码程序 > datastoreio/+matlab/+io/+datastore/@ImageDatastore/splitEachLabel.m
function varargout = splitEachLabel(imds, varargin)
%SPLITEACHLABEL Split each label with specified proportions of files.
% [DS1,DS2] = SPLITEACHLABEL(DS,F1) splits the datastore's files from
% each label according to proportion F1. DS1 contains the files
% corresponding to F1 and DS2 contains the remaining files.
% F1 can be:
% - A fraction such that 0 < F1 < 1
% - A numerical value denoting the number of files in each label,
% where F1 >= 1
%
% [DS1,...,DSm] = SPLITEACHLABEL(DS,F1,...,Fn) splits the datastore's
% files from each label according to the proportions F1,...,Fn into
% datastores DS1,...DSm, where m = n + 1.
% The proportions F1,...,Fn must satisfy either of the following:
% - The sum of fractions F1,...,Fn must be less than or equal to 1.
% - The sum of numbers F1,...,Fn must be less than or equal to the
% minimum of the number of files in each label.
%
% [DS1,...,DSm] = SPLITEACHLABEL(__,'randomized') uses randperm to draw
% image files randomly using randperm from each label according to
% proportions F1,...,Fn resulting in datastores DS1,...DSm, where
% m = n + 1.
%
% [DS1,...,DSm] = SPLITEACHLABEL(__,'Include',LABELS) includes image
% files belonging to LABELS. LABELS and the Labels property must have the
% same type.
%
% [DS1,...,DSm] = SPLITEACHLABEL(__,'Exclude',LABELS) excludes image
% files belonging to LABELS. LABELS and the Labels property must have the
% same type.
%
% Note: The 'Include' and 'Exclude' parameters cannot be combined.
%
% Example: Split using percentages
% --------------------------------
% folders = fullfile(matlabroot,'toolbox','matlab',{'demos','imagesci'});
% exts = {'.jpg','.png','.tif'};
% imds = imageDatastore(folders,'LabelSource','foldernames','FileExtensions',exts)
%
% % Split 60% of the files from each label into ds60 and the rest into dsRest
% [ds60,dsRest] = splitEachLabel(imds,0.6)
%
% % split 70% of the files from each label into ds70, 20% into ds20 and the rest into dsRest
% [ds70,ds20,dsRest] = splitEachLabel(imds,0.7,0.2)
%
% Example: Split using number of files
% ------------------------------------
% folders = fullfile(matlabroot,'toolbox','matlab',{'demos','imagesci'});
% exts = {'.jpg','.png','.tif'};
% imds = imageDatastore(folders,'LabelSource','foldernames','FileExtensions',exts)
%
% % Use countEachLabel to find the number of files in each label
% tbl = countEachLabel(imds);
%
% % Split using the minimum from the number of files in each label
% [dsMin,dsRest] = splitEachLabel(imds,min(tbl.Count))
%
% Example: Split by drawing random images
% ---------------------------------------
% folders = fullfile(matlabroot,'toolbox','matlab',{'demos','imagesci'});
% exts = {'.jpg','.png','.tif'};
% imds = imageDatastore(folders,'LabelSource','foldernames','FileExtensions',exts)
%
% % Draw 70% of files from each label randomly into randDs70 with the rest in randDsRest
% [randDs70,randDsRest] = splitEachLabel(imds,0.7,'randomized')
%
% Example: Include or Exclude specific labels
% -------------------------------------------
% folders = fullfile(matlabroot,'toolbox','matlab',{'demos','imagesci'});
% exts = {'.jpg','.png','.tif'};
% imds = imageDatastore(folders,'LabelSource','foldernames','FileExtensions',exts)
%
% % Split by drawing 70% of the images randomly from only files with the demos label
% [dsDemos,dsDemosRest] = splitEachLabel(imds,0.7,'randomized','Include','demos')
%
% % Split by drawing 70% of the images randomly but exclude files with the imagesci label
% [dsDemos,dsDemosRest] = splitEachLabel(imds,0.7,'randomized','Exclude','imagesci')
%
% See also imageDatastore, countEachLabel, shuffle, hasdata, readimage,
% readall, preview, reset.
% Copyright 2015 The MathWorks, Inc.
try
if isempty(imds.Labels)
error(message('MATLAB:datastoreio:imagedatastore:splitWhenLabelsEmpty'));
end
numargin = nargin - 1;
[idx, nonNum] = getNumericIndex(numargin, varargin);
options = '';
if nonNum
options = validatestring(varargin{idx}, {'randomized', 'Include', 'Exclude'}, 'datastoreio:imagedatastore:splitEachLabel');
nums = [varargin{1:idx-1}];
else
nums = [varargin{1:idx}];
end
numargout = nargout;
if numargout == 0
numargout = 1;
end
numsGiven = numel(nums);
if numargout > numsGiven + 1
error(message('MATLAB:datastoreio:imagedatastore:numOutMoreThanNumProps', numsGiven + 1));
end
if any(nums <= 0)
error(message('MATLAB:datastoreio:imagedatastore:nonPositiveProportions'));
end
randomized = false;
if strcmpi(options, 'randomized')
idx = idx + 1;
randomized = true;
elseif numargin > idx + 1
try
options = validatestring(varargin{idx+2}, {'Include', 'Exclude'});
catch exc
error(message('MATLAB:datastoreio:imagedatastore:optionsAfterIncExc'));
end
end
includeAll = true;
includeNonDefault = false;
if idx < numargin
inpP = inputParser;
addParameter(inpP, 'Include', {});
addParameter(inpP, 'Exclude', {});
inpP.FunctionName = 'splitEachLabel';
parse(inpP, varargin{idx:end});
if isempty(inpP.UsingDefaults)
error(message('MATLAB:datastoreio:imagedatastore:combineIncExcNVPair'));
end
res = inpP.Results;
if ~ismember('Exclude', inpP.UsingDefaults)
includeAll = false;
include = res.Exclude;
if ischar(include)
include = {include};
end
errorBasedOnType(imds.Labels, include);
include = setdiff(imds.Labels, include);
if iscategorical(res.Exclude) && any(isundefined(res.Exclude))
include = include(~isundefined(include));
end
elseif ~ismember('Include', inpP.UsingDefaults)
includeAll = false;
includeNonDefault = true;
include = res.Include;
if ischar(include)
include = {include};
end
errorBasedOnType(imds.Labels, include);
end
if numel(include) == imds.NumFiles && isempty(setdiff(imds.Labels, include))
includeAll = true;
end
end
if ~includeAll
idxes = compareLabels(imds.Labels, include, imds.NumFiles, includeNonDefault);
idxes = find(idxes);
copiedDs = getNewDs(imds, idxes);
else
copiedDs = copy(imds);
end
if isempty(copiedDs.Files)
varargout = arrayfun(@(x)copy(copiedDs), 1:numargout, 'UniformOutput', false);
return;
end
[uniq, count, groups, nans] = groupAndCountLabels(copiedDs);
numLabels = numel(uniq);
fileCount = getFileCount(numargout, nums, count, numLabels);
res = getLabelIndexes(groups, numLabels, nans, randomized);
for ii = 2:numargout+1
indexes = [];
% If the number of files from each label is zero, then
% output an empty datastore. For example, if each label
% contains less than 5 files and splitEachLabel is as below:
% [ds1, ds2, ds3] = splitEachLabel(ds, 0.1, 0.1);
% ds1 and ds2 will be empty; ds3 will contain all the files.
if nnz(fileCount(:,ii) - fileCount(:, ii-1)) == 0
varargout{ii-1} = imageDatastore({});
continue;
end
for j = 1:numLabels
startIdx = fileCount(j, ii-1) + 1;
endIdx = fileCount(j, ii);
r = res{j};
idxes = r(startIdx:endIdx);
indexes = [indexes; idxes];
end
% Maintain indexes of original datastore
indexes = sort(indexes);
varargout{ii-1} = getNewDs(copiedDs, indexes);
end
catch e
throw(e)
end
end
function fileCount = getFileCount(numargout, nums, count, numLabels)
lessThanOne = nums < 1;
if all(lessThanOne)
numsum = sum(nums);
if numsum > 1
error(message('MATLAB:datastoreio:imagedatastore:sumFracMoreThanOne'));
end
if numargout <= numel(nums)
nums = nums(1:numargout);
end
fileCount = round(count * nums);
restCount = count - sum(fileCount, 2);
negativeRest = find(restCount < 0);
if ~isempty(negativeRest);
fileCount(negativeRest, end) = fileCount(negativeRest, end) + restCount(negativeRest);
restCount(negativeRest) = 0;
end
elseif all(~lessThanOne)
if ~isequal(round(nums), nums)
error(message('MATLAB:datastoreio:imagedatastore:fracMoreThanOne'));
end
numsum = sum(nums);
if numsum > min(count)
error(message('MATLAB:datastoreio:imagedatastore:sumPropMoreThanLabels', numsum, min(count)));
end
if numargout <= numel(nums)
nums = nums(1:numargout);
end
fileCount = repmat(nums, numLabels, 1);
restCount = count - sum(fileCount, 2);
else
error(message('MATLAB:datastoreio:imagedatastore:neitherFracNorNum'));
end
fileCount = cumsum([fileCount, restCount], 2);
fileCount = [zeros(size(count)), fileCount];
end
function [idx, nonNumFound] = getNumericIndex(numargs, varargs)
oneNumeric = false;
nonNumFound = false;
if numargs < 1
error(message('MATLAB:datastoreio:imagedatastore:emptyArguments'));
end
for idx = 1:numargs
v = varargs{idx};
if ~isnumeric(v)
nonNumFound = true;
break;
end
if ~isscalar(v)
error(message('MATLAB:datastoreio:imagedatastore:nonScalarNumericalInput'));
end
oneNumeric = true;
end
if ~oneNumeric
error(message('MATLAB:datastoreio:imagedatastore:noNumericalInput'));
end
end
function ds = getNewDs(ds, idxes)
if isempty(idxes)
ds = imageDatastore({});
return;
end
ds = copy(ds);
% Signature for initFromReadFcn:
% initFromReadFcn(ds, readFcn, files, resolved, includeSubfolders)
%
% - resolved = false. Files are already resolved so we don't lookup
% the path and verify the existence of files
% - includeSubfolders = false. Files are resolved and none of them
% are folders.
initFromReadFcn(ds, ds.ReadFcn, ds.Files(idxes), true, false);
ds.Labels = ds.Labels(idxes);
setExistingCategories(ds);
end
function res = getLabelIndexes(g, numLabels, nans, randomized)
res = cell(numLabels, 1);
if ~isempty(nans)
res{end} = find(nans);
numLabels = numLabels - 1;
end
for i = 1:numLabels
indexes = find(g==i);
if randomized
indexes = indexes(randperm(numel(indexes)));
end
res{i} = indexes;
end
end
function idxes = compareLabels(labels, other, numLabels, includeNonDefault)
idxes = zeros(numLabels, 1);
% Take only the unique labels to compare.
other = unique(other);
compareFcn = @(x)labels == x;
switch class(labels)
case 'cell'
compareFcn = @(x)strcmp(labels, x{1});
case 'categorical'
switch class(other)
case 'cell'
compareFcn = @(x)labels == x{1};
case 'categorical'
% if other says to include undefined categoricals
% include them.
undef = isundefined(other);
if any(undef)
other = other(~undef);
compares = isundefined(labels);
errorIfIncludeNotExists(includeNonDefault, compares, categorical(nan));
idxes = or(idxes, compares);
end
end
end
for ii = 1:numel(other)
element = other(ii);
compares = compareFcn(element);
errorIfIncludeNotExists(includeNonDefault, compares, element);
idxes = or(idxes, compares);
end
end
function errorIfIncludeNotExists(includeNonDefault, compares, other)
if ~includeNonDefault || nnz(compares) ~= 0
return;
end
switch class(other)
case 'cell'
other = other{1};
case 'categorical'
other = char(other);
otherwise
other = mat2str(other);
end
error(message('MATLAB:datastoreio:imagedatastore:includeNotInLabels', other));
end
function errorBasedOnType(labels, other)
switch class(labels)
case 'cell'
if iscellstr(other)
return;
end
case 'categorical'
switch class(other)
case 'cell'
if iscellstr(other)
return;
end
case 'categorical'
return;
end
case 'logical'
if islogical(other)
return;
end
otherwise
if isnumeric(other)
return;
end
end
error(message('MATLAB:datastoreio:imagedatastore:incExcInvalidType'));
end