augmentedI​mageDatast​ore内のプロパティ​について

1 view (last 30 days)
Ren Okumura
Ren Okumura on 31 Oct 2022
Commented: Ren Okumura on 2 Nov 2022
ラベル付きのイメージデータストアをaugmentedImageDatastoreを使用してoutputSizeを変更しました。
その拡張イメージデータストアのプロパティを確認すると、元のイメージデータストアに付いていたラベルが確認できません。
この拡張イメージデータストアに、ラベルの情報は含まれているのでしょうか。
教えていただけると幸いです。よろしくお願いいたします。
  3 Comments
Show 1 older comment
Ren Okumura
Ren Okumura on 1 Nov 2022
Moved: Atsushi Ueno on 1 Nov 2022
教えていただきありがとうございます。
分からず悩んでいた点だったため、大変助かりました！
Atsushi Ueno
Atsushi Ueno on 1 Nov 2022
先の回答は誤り（でもないが不的確な回答）なので、コメントに移動しました。

Sign in to comment.

Sign in to answer this question.

Accepted Answer

Atsushi Ueno
Atsushi Ueno on 1 Nov 2022
Ran in:
augmentedImageDatastore center crop does not return datastore with labels - MATLAB Answers - MATLAB Central (mathworks.com)
augmentedImageDatastore からデータを読み取る - MATLAB read - MathWorks 日本
>この拡張イメージデータストアに、ラベルの情報は含まれているのでしょうか。
⇒はい、augmentedImageDatastore に ラベル情報は存在します。それはprivate属性の為ワークスペースに表示されません。augmentedImageDatastore の read 関数により取得する事が出来ます。
[XTrain,YTrain] = digitTrain4DArrayData; % XTrainは手書き文字5000枚、YTrainはラベル情報
augimds = augmentedImageDatastore([28 28 1],XTrain,YTrain) % augimdsのプロパティにラベル情報は無い(見えない)
augimds =
augmentedImageDatastore with properties: NumObservations: 5000 MiniBatchSize: 128 DataAugmentation: 'none' ColorPreprocessing: 'none' OutputSize: [28 28] OutputSizeMode: 'resize' DispatchInBackground: 0
read(augimds) % Read関数でラベル情報(response:この例は手書き文字画像の種類)が得られた
ans = 128×2 table
input response ______________ ________ {28×28 double} 3 {28×28 double} 3 {28×28 double} 7 {28×28 double} 7 {28×28 double} 7 {28×28 double} 7 {28×28 double} 9 {28×28 double} 9 {28×28 double} 7 {28×28 double} 7 {28×28 double} 3 {28×28 double} 9 {28×28 double} 9 {28×28 double} 7 {28×28 double} 3 {28×28 double} 7
  3 Comments
Show 1 older comment
Atsushi Ueno
Atsushi Ueno on 1 Nov 2022
Ran in:
定義元のソースコードを読むのが一番ですが、場合によっては隠ぺいされていて読めなかったり、読めても無茶苦茶長くて挫折するので、上記の「構造体にコピー」する方法は有用だと思います。
type augmentedImageDatastore.m
% augmentedImageDatastore Generate batches of augmented image data % % ds = augmentedImageDatastore(outputSize,imds) returns an % augmentedImageDatastore. outputSize is a two element vector which % specifies the output image size in the form [outputWidth, % outputHeight]. imds is an imageDatastore. % % ds = augmentedImageDatastore(outputSize,X,Y) returns an % augmentedImageDatastore given arrays X and Y that define examples and % corresponding responses. % % ds = augmentedImageDatastore(outputSize,X) returns an augmentedImageDatastore % given an input array X that defines examples. % % ds = augmentedImageDatastore(outputSize,tbl) returns a % an augmentedImageDatastore given a tbl which contains predictors in the % first column as either absolute or relative image paths or images. % If responses are specified, responses must be in the second column as % categorical labels for the images. In a regression problem, responses % must be in the second column as either vectors or cell arrays containing % 3-D arrays or in multiple columns as scalars. % % ds = augmentedImageDatastore(outputSize,tbl, responseName,___) % returns an augmentedImageDatastore which yields predictors and responses. % tbl is a MATLAB table. responseName is a character vector specifying % the name of the variable in tbl that contains the responses. % % ds = augmentedImageDatastore(outputSize,tbl,responseNames,___) returns % an augmentedImageDatastore for use in multi-output regression problems. % tbl is a MATLAB table. responseNames is a cell array of character % vectors specifying the names of the variables in tbl that contain the % responses. % % ds = augmentedImageDatastore(___,Name,Value) returns an % augmentedImageDatastore using Name/Value pairs to configure % image-preprocessing options. % % Parameters include: % % 'ColorPreprocessing' A scalar string or character vector specifying % color channel pre-processing. This option can % be used when you have a training set that % contains both color and grayscale image data % and you need data created by the datastore to % be strictly color or grayscale. Options are: % 'gray2rgb','rgb2gray','none'. For example, if % you need to train a network that expects color % images but some of the images in your training % set are grayscale, then specifying the option % 'gray2rgb' will replicate the color channels of % the grayscale images in the input image set to % create MxNx3 output images. % % Default: 'none' % % 'DataAugmentation' A scalar imageDataAugmenter object, string, or % character array that specifies % the kinds of image data augmentation that will % be applied to generated images. % % Default: 'none' % % 'DispatchInBackground' Accelerate image augmentation by asyncronously % reading, augmenting, and queueing augmented % images for use in training. Requires Parallel % Computing Toolbox. % % Default: false % % 'OutputSizeMode' A scalar string or character vector specifying the % technique used to adjust image sizes to the % specified 'OutputSize'. Options are: 'resize', % 'centercrop', 'randcrop'. % % Default: 'resize' % % augmentedImageDatastore Properties: % % ColorPreprocessing - Defines color channel manipulation % DataAugmentation - Defines data augmentation used % DispatchInBackground - Whether background dispatch is used % MiniBatchSize - Number of images returned in each read % NumObservations - Total number of images in an epoch % OutputSize - Vector of [width,height] of output images % OutputSizeMode - Method used to resize output images % % augmentedImageDatastore Methods: % % augmentedImageDatastore - Construct an augmentedImageDatastore % hasdata - Returns true if there is more data in the datastore % partitionByIndex - Partitions an augmentedImageDatastore given indices % partition - Returns a new datastore that represents a single % partitioned portion of the original datastore % numpartitions - Returns an estimate for a reasonable number of % partitions to use with the partition function, % according to the total data size % subset - Subsets the datastore according to the specified % file indices % preview - Reads the first image from the datastore % read - Reads a MiniBatch of data from the datastore % readall - Reads all observations from the datastore % readByIndex - Random access read from datastore given indices % reset - Resets datastore to the start of the data % shuffle - Shuffles the observations in the datastore % transform - Create an altered form of the current datastore by % specifying a function handle that will execute % after read on the current datastore. % combine - Create a new datastore that horizontally % concatenates the result of read from two or more % input datastores. % % Example 1 % --------- % Train a convolutional neural network on some synthetic images of % handwritten digits. Apply random rotations during training to add % rotation invariance to trained network. % % [XTrain, YTrain] = digitTrain4DArrayData; % % imageSize = [28 28 1]; % % layers = [ ... % imageInputLayer(imageSize, 'Normalization', 'none'); % convolution2dLayer(5,20); % reluLayer(); % maxPooling2dLayer(2,'Stride',2); % fullyConnectedLayer(10); % softmaxLayer(); % classificationLayer()]; % % opts = trainingOptions('sgdm', 'Plots', 'training-progress'); % % imageAugmenter = imageDataAugmenter('RandRotation',[-10 10]); % % ds = augmentedImageDatastore(imageSize,XTrain,YTrain,'DataAugmentation',imageAugmenter); % % net = trainNetwork(ds, layers, opts); % % Example 2 % --------- % Resize all images in imageDatastore during inference to the required % size of the inputImageLayer of alexnet. % % net = alexnet; % imds = imageDatastore(fullfile(matlabroot,'toolbox','matlab','imagesci','peppers.png')); % imds.Files = repelem(imds.Files,5,1); % ds = augmentedImageDatastore(net.Layers(1).InputSize(1:2),imds); % Ypred = classify(net,ds) % See also imageDataAugmenter, imageInputLayer, trainNetwork % Copyright 2017-2022 The MathWorks, Inc. classdef augmentedImageDatastore <... matlab.io.Datastore &... matlab.io.datastore.MiniBatchable &... matlab.io.datastore.BackgroundDispatchable &... matlab.io.datastore.PartitionableByIndex &... matlab.io.datastore.internal.RandomizedReadable &... matlab.io.datastore.mixin.Subsettable &... matlab.mixin.CustomDisplay &... matlab.mixin.Copyable properties (SetAccess = protected, Dependent) %NumObservations - Number of observations % % The number of observations in the datastore. NumObservations %Files - Contained files in Datastore % % Cell array of character vectors of image files. Files end properties (Dependent) %AlternativeFileSystemRoots % % Alternate file system root paths for the Files. AlternateFileSystemRoots %MiniBatchSize - MiniBatch Size % % The number of observations returned as rows in the table % returned by the read method. MiniBatchSize end properties (Access = private) DatastoreInternal ImageAugmenter end properties (SetAccess = private) %DataAugmentation - Augmentation applied to input images % % DataAugmentation is a scalar imageDataAugmenter object or a % character vector or string. When DataAugmentation is 'none' % no augmentation is applied to input images. DataAugmentation %ColorPreprocessing - Pre-processing of input image color channels % % ColorPreprocessing is a character vector or string specifying % pre-proprocessing operations performed on color channels of % input images. This property is used to ensure that all output % images from the datastore have the number of color channels % required by inputImageLayer. Valid values are % 'gray2rgb','rgb2gray', and 'none'. If an input images already % has the desired number of color channels, no operation is % performed. For example, if 'gray2rgb' is specified and an % input image already has 3 channels, no operation is performed. ColorPreprocessing %OutputSize - Size of output images % % OutputSize is a two element numeric vector of the form % [numRows, numColumns] that specifies the size of output images % returned by augmentedImageDatastore. OutputSize %OutputSizeMode - Method used to resize output images. % % OutputSizeMode is a character vector or string specifying the % method used to resize output images to the requested % OutputSize. Valid values are 'centercrop', 'randcrop', and % 'resize' (default). OutputSizeMode end properties (Access = private) OutputRowsColsChannels % The expected output image size [numRows, numCols, numChannels]. end methods(Access = protected) % Override copyElement method: function cpObj = copyElement(obj) cpObj = copyElement@matlab.mixin.Copyable(obj); % Make a deep copy of the underlying Datastore cpObj.DatastoreInternal = copy(obj.DatastoreInternal); end end methods function self = augmentedImageDatastore(varargin) narginchk(2,inf) inputs = self.parseInputs(varargin{:}); if self.NumObservations > 0 self.ImageAugmenter = inputs.DataAugmentation; self.DispatchInBackground = inputs.DispatchInBackground; reset(self.DatastoreInternal); end end function set.MiniBatchSize(self,batchSize) self.DatastoreInternal.MiniBatchSize = batchSize; end function batchSize = get.MiniBatchSize(self) batchSize = self.DatastoreInternal.MiniBatchSize; end function files = get.Files(self) if isa(self.DatastoreInternal,'nnet.internal.cnn.datastore.ImageDatastoreMiniBatchDatastore') files = self.DatastoreInternal.imds.Files; else files = {}; end end function altRoots = get.AlternateFileSystemRoots(self) if isa(self.DatastoreInternal,'nnet.internal.cnn.datastore.ImageDatastoreMiniBatchDatastore') altRoots = self.DatastoreInternal.imds.AlternateFileSystemRoots; else altRoots = {}; end end function set.AlternateFileSystemRoots(self,newRoots) if isa(self.DatastoreInternal,'nnet.internal.cnn.datastore.ImageDatastoreMiniBatchDatastore') self.DatastoreInternal.imds.AlternateFileSystemRoots = newRoots; else error(message('nnet_cnn:augmentedImageDatastore:invalidSetAltRoot')); end end function numObs = get.NumObservations(self) numObs = self.DatastoreInternal.NumObservations; end end methods function reset(self) self.DatastoreInternal.reset(); end function [data,info] = readByIndex(self,indices) [input,info] = self.DatastoreInternal.readByIndex(indices); input = self.applyAugmentationPipelineToBatch(input); [data,info] = iConvertDatastoreOutputToTable(input,info); end function [data,info] = read(self) if ~self.hasdata() error(message('nnet_cnn:augmentedImageDatastore:outOfData')); end [input,info] = self.DatastoreInternal.read(); input = self.applyAugmentationPipelineToBatch(input); [data,info] = iConvertDatastoreOutputToTable(input,info); end function TF = hasdata(self) TF = hasdata(self.DatastoreInternal); end function newds = partitionByIndex(self,indices) newds = copy(self); newds.DatastoreInternal = partitionByIndex(self.DatastoreInternal,indices); end function newds = subset(self,indices) newds = partitionByIndex(self,indices); end function n = numobservations(self) n = self.NumObservations; end end methods (Hidden) % Reorder is an undocumented interface for doing in place shuffling % without a copy. function reorder(self,indices) self.DatastoreInternal.reorder(indices); end function frac = progress(self) frac = progress(self.DatastoreInternal); end end methods (Access = private) function determineExpectedOutputSize(self,X) % If a user specifies a ColorPreprocessing option, we know the % number of channels to expect in each mini-batch. If they % don't specify a ColorPreprocessing option, we need to look at % an example from the underlying datastore and assume all % images will have a consistent number of channels when forming % mini-batches. if strcmp(self.ColorPreprocessing,'rgb2gray') self.OutputRowsColsChannels = [self.OutputSize,1]; elseif strcmp(self.ColorPreprocessing,'gray2rgb') self.OutputRowsColsChannels = [self.OutputSize,3]; elseif strcmp(self.ColorPreprocessing,'none') if iscell(X) X = X{1}; end exampleNumChannels = size(X,3); self.OutputRowsColsChannels = [self.OutputSize,exampleNumChannels]; else assert(false,'Unexpected ColorPreprocessing option.'); end end function Xout = applyAugmentationPipelineToBatch(self,X) if iscell(X) Xout = cellfun(@(c) self.applyAugmentationPipeline(c),X,'UniformOutput',false); else batchSize = size(X,4); Xout = cell(batchSize,1); for obs = 1:batchSize temp = self.preprocessColor(X(:,:,:,obs)); temp = self.augmentData(temp); Xout{obs} = self.resizeData(temp); end end end function Xout = applyAugmentationPipeline(self,X) if isempty(self.OutputRowsColsChannels) self.determineExpectedOutputSize(X); end if isequal(self.ColorPreprocessing,'none') && (size(X,3) ~= self.OutputRowsColsChannels(3)) error(message('nnet_cnn:augmentedImageDatastore:mismatchedNumberOfChannels','''ColorPreprocessing''')); end temp = self.preprocessColor(X); temp = self.augmentData(temp); Xout = self.resizeData(temp); end function miniBatchData = augmentData(self,miniBatchData) if ~strcmp(self.DataAugmentation,'none') miniBatchData = self.ImageAugmenter.augment(miniBatchData); end end function Xout = resizeData(self,X) inputSize = size(X); if isequal(inputSize(1:2),self.OutputSize) Xout = X; % no-op if X is already desired Outputsize return end if strcmp(self.OutputSizeMode,'resize') Xout = augmentedImageDatastore.resizeImage(X,self.OutputSize); elseif strcmp(self.OutputSizeMode,'centercrop') Xout = augmentedImageDatastore.centerCrop(X,self.OutputSize); elseif strcmp(self.OutputSizeMode,'randcrop') Xout = augmentedImageDatastore.randCrop(X,self.OutputSize); end end function Xout = preprocessColor(self,X) if strcmp(self.ColorPreprocessing,'rgb2gray') Xout = convertRGBToGrayscale(X); elseif strcmp(self.ColorPreprocessing,'gray2rgb') Xout = convertGrayscaleToRGB(X); elseif strcmp(self.ColorPreprocessing,'none') Xout = X; end end end methods (Access = 'protected') function groups = getPropertyGroups(self) groups = getPropertyGroups@matlab.mixin.CustomDisplay(self); if ~isa(self.DatastoreInternal,'nnet.internal.cnn.datastore.ImageDatastoreMiniBatchDatastore') % Don't display Files or AlternativeFileSystemRoots if the % underlying data was not specified as an ImageDatastore. currentStruct = groups.PropertyList; newStruct = rmfield(currentStruct,'Files'); newStruct = rmfield(newStruct,'AlternateFileSystemRoots'); groups = matlab.mixin.util.PropertyGroup(newStruct); end end end methods (Access = 'private') function inputStruct = parseInputs(self,varargin) p = inputParser(); p.addRequired('outputSize',@outputSizeValidator); p.addRequired('X'); p.addOptional('Y',[]); p.addParameter('DataAugmentation','none',@augmentationValidator); colorPreprocessing = 'none'; p.addParameter('ColorPreprocessing','none',@colorPreprocessingValidator); outputSizeMode = 'resize'; p.addParameter('OutputSizeMode','resize',@outputSizeModeValidator); backgroundExecutionValidator = @(TF) validateattributes(TF,... {'numeric','logical'},{'scalar','real'},mfilename,'BackgroundExecution'); p.addParameter('DispatchInBackground',false,backgroundExecutionValidator); p.addParameter('BackgroundExecution',false,backgroundExecutionValidator); responseNames = []; if (istable(varargin{2}) && ~isempty(varargin{2})) tbl = varargin{2}; if (length(varargin) > 2) && (ischar(varargin{3}) || isstring(varargin{3}) || iscellstr(varargin{3})) if checkValidResponseNames(varargin{3},tbl) responseNames = varargin{3}; varargin(3) = []; end end end p.parse(varargin{:}); inputStruct = manageDispatchInBackgroundNameValue(p); self.DataAugmentation = inputStruct.DataAugmentation; self.OutputSize = inputStruct.outputSize(1:2); self.OutputSizeMode = outputSizeMode; self.ColorPreprocessing = colorPreprocessing; % Check if Y was specified for table or imageDatastore inputs. propertiesWithDefaultValues = string(p.UsingDefaults); if (isa(inputStruct.X,'matlab.io.datastore.ImageDatastore') || isa(inputStruct.X,'table')) && ~any(propertiesWithDefaultValues == "Y") error(message('nnet_cnn:augmentedImageDatastore:invalidYSpecification',class(inputStruct.X))); end if ~isempty(responseNames) inputStruct.X = selectResponsesFromTable(inputStruct.X,responseNames); inputStruct.Y = responseNames; end % Validate numeric inputs if isnumeric(inputStruct.X) || iscategorical(inputStruct.X) || islogical(inputStruct.X) validateattributes(inputStruct.X,{'single','double','logical','categorical','uint8','int8','uint16','int16','uint32','int32'},... {'nonsparse','real'},mfilename,'X'); validateattributes(inputStruct.Y,{'single','double','logical','categorical','uint8','int8','uint16','int16','uint32','int32','categorical'},... {'nonsparse'},mfilename,'Y'); if ~isempty(inputStruct.Y) if iscategorical(inputStruct.Y) if ~isvector(inputStruct.Y) error(message('nnet_cnn:augmentedImageDatastore:InvalidClassificationResponse')) else % Make sure responses are stored as column % vectors inputStruct.Y = inputStruct.Y(:); end end end end try self.DatastoreInternal = nnet.internal.cnn.datastore.MiniBatchDatastoreFactory.createMiniBatchDatastore(... inputStruct.X ,inputStruct.Y, inputStruct.outputSize); catch ME throwAsCaller(ME); end function TF = colorPreprocessingValidator(sIn) colorPreprocessing = validatestring(sIn,{'none','rgb2gray','gray2rgb'},... mfilename,'ColorPreprocessing'); TF = true; end function TF = outputSizeModeValidator(sIn) outputSizeMode = validatestring(sIn,... {'resize','centercrop','randcrop'},mfilename,'OutputSizeMode'); TF = true; end function TF = outputSizeValidator(sizeIn) validateattributes(sizeIn,... {'numeric'},{'vector','integer','finite','nonsparse','real','positive'},mfilename,'OutputSize'); if (numel(sizeIn) ~= 2) && (numel(sizeIn) ~=3) error(message('nnet_cnn:augmentedImageDatastore:invalidOutputSize')); end TF = true; end end end methods(Static, Hidden = true) function self = loadobj(S) self = augmentedImageDatastore(S.OutputSize,S.DatasourceInternal,... 'DispatchInBackground',S.BackgroundExecution,... 'ColorPreprocessing',S.ColorPreprocessing,... 'DataAugmentation',S.DataAugmentation,... 'OutputSizeMode',S.OutputSizeMode); if isfield(S,'MiniBatchSize') self.MiniBatchSize = S.MiniBatchSize; else self.MiniBatchSize = 128; end end end methods (Hidden) function S = saveobj(self) S = struct('BackgroundExecution',self.DispatchInBackground,... 'ColorPreprocessing',self.ColorPreprocessing,... 'DataAugmentation',self.DataAugmentation,... 'OutputSize',self.OutputSize,... 'OutputSizeMode',self.OutputSizeMode,... 'DatasourceInternal',self.DatastoreInternal,... 'MiniBatchSize',self.MiniBatchSize); end end methods (Hidden, Static) function imOut = resizeImage(im,outputSize) classIn = class(im); interpMethod = 'linear'; if isa(im,'categorical') cats = categories(im); im = single(im); interpMethod = 'nearest'; elseif islogical(im) im = uint8(im); interpMethod = 'nearest'; end ippResizeSupportedWithCast = isa(im,'int8') || isa(im,'uint16') || isa(im,'int16'); ippResizeSupportedForType = isa(im,'uint8') || isa(im,'single'); ippResizeSupported = ippResizeSupportedWithCast || ippResizeSupportedForType; if ippResizeSupportedWithCast im = single(im); end if ippResizeSupported imOut = nnet.internal.cnnhost.resizeImage2D(im,outputSize,interpMethod,true); else if string(interpMethod) == "linear" interpMethod = 'bilinear'; end imOut = imresize(single(im),'OutputSize',outputSize,'Method',interpMethod); end classIn = string(classIn); if classIn == "categorical" imOut = categorical(imOut,1:numel(cats),cats); elseif classIn == "logical" imOut = logical(imOut); end end function im = centerCrop(im,outputSize) sizeInput = size(im); if any(sizeInput(1:2) < outputSize) error(message('nnet_cnn:augmentedImageDatastore:invalidCropOutputSize','''OutputSizeMode''',mfilename, '''centercrop''','''OutputSize''')); end x = (size(im,2) - outputSize(2)) / 2; y = (size(im,1) - outputSize(1)) / 2; im = augmentedImageDatastore.crop(im,... [x y, outputSize(2), outputSize(1)]); end function rect = randCropRect(im,outputSize) x = randi([1, min(size(im,2),max(1,size(im,2)-outputSize(2)))]); y = randi([1, min(size(im,1),max(1,size(im,1)-outputSize(1)))]); rect = [x y outputSize(2), outputSize(1)]; end function rect = randCropCuboid(im,outputSize) x = randi([1, min(size(im,2),max(1,size(im,2)-outputSize(2)))]); y = randi([1, min(size(im,1),max(1,size(im,1)-outputSize(1)))]); z = randi([1, min(size(im,3),max(1,size(im,3)-outputSize(3)))]); rect = [x y z outputSize(2), outputSize(1), outputSize(3)]; end function im = randCrop(im,outputSize) sizeInput = size(im); if any(sizeInput(1:2) < outputSize) error(message('nnet_cnn:augmentedImageDatastore:invalidCropOutputSize','''OutputSizeMode''',mfilename, '''randcrop''','''OutputSize''')); end rect = augmentedImageDatastore.randCropRect(im,outputSize); im = augmentedImageDatastore.cropGivenDiscreteValuedRect(im,rect); end function B = cropGivenDiscreteValuedRect(A,rect) rowStart = rect(2); rowEnd = rowStart + rect(4) - 1; colStart = rect(1); colEnd = colStart + rect(3) - 1; B = A(rowStart:rowEnd,colStart:colEnd,:); end function B = cropGivenDiscreteValuedCuboid(A,cuboid) rowStart = cuboid(2); rowEnd = rowStart + cuboid(5) - 1; colStart = cuboid(1); colEnd = colStart + cuboid(4) - 1; depthStart = cuboid(3); depthEnd = depthStart + cuboid(6) - 1; B = A(rowStart:rowEnd,colStart:colEnd,depthStart:depthEnd,:); end function B = crop(A,rect) % rect is [x y width height] in floating point. % Convert from (x,y) real coordinates to [m,n] indices. rect = floor(rect); m1 = rect(2) + 1; m2 = rect(2) + rect(4); n1 = rect(1) + 1; n2 = rect(1) + rect(3); m1 = min(size(A,1),max(1,m1)); m2 = min(size(A,1),max(1,m2)); n1 = min(size(A,2),max(1,n1)); n2 = min(size(A,2),max(1,n2)); B = A(m1:m2, n1:n2, :, :); end end end function TF = checkValidResponseNames(responseNames, tbl) % iAssertValidResponseNames Assert that the response names are variables % of the table and they do not refer to the first column. variableNames = tbl.Properties.VariableNames; refersToFirstColumn = ismember( variableNames(1), responseNames ); responseNamesAreAllVariables = all( ismember(responseNames,variableNames) ); TF = ~(refersToFirstColumn || ~responseNamesAreAllVariables); end function resTbl = selectResponsesFromTable(tbl, responseNames) % iSelectResponsesFromTable Return a new table with only the first column % (predictors) and the variables specified in responseNames. variableNames = tbl.Properties.VariableNames; varTF = ismember(variableNames, responseNames); % Make sure to select predictors (first column) as well varTF(1) = 1; resTbl = tbl(:,varTF); end function TF = augmentationValidator(valIn) if ischar(valIn) || isstring(valIn) TF = string('none').contains(lower(valIn)); %#ok<STRQUOT> elseif isa(valIn,'imageDataAugmenter') && isscalar(valIn) TF = true; else TF = false; end end function im = convertRGBToGrayscale(im) if (ndims(im) == 3 && size(im,3) == 3) im = rgb2gray(im); end end function im = convertGrayscaleToRGB(im) if size(im,3) == 1 im = repmat(im,[1 1 3]); end end function c = iFormatResponse(X) % Format the responses. If the responses were numeric, return as a cell % array with each observation in a cell. if isnumeric(X) if isvector(X) % Put each element in a cell c = num2cell(X); else % Always assume that the observation dimension is 4 obsDim = 4; % Put each observation in a cell c = num2cell(X,1:(obsDim-1)); end % Force c as column. c = c(:); else % Response is already a cell or categorical. c = X; end end function [data,info] = iConvertDatastoreOutputToTable(input,info) % Process the output of the datastore and create the table returned when % calling "read". response = info.Response; info = rmfield(info,'Response'); if isempty(response) data = table(input); else response = iFormatResponse(response); data = table(input,response); end end function resultsStruct = manageDispatchInBackgroundNameValue(p) resultsStruct = p.Results; DispatchInBackgroundSpecified = ~any(strncmp('DispatchInBackground',p.UsingDefaults,length('DispatchInBackground'))); BackgroundExecutionSpecified = ~any(strncmp('BackgroundExecution',p.UsingDefaults,length('BackgroundExecution'))); % In R2017b, BackgroundExecution was name used to control % DispatchInBackground. Allow either to be specified. if BackgroundExecutionSpecified && ~DispatchInBackgroundSpecified resultsStruct.DispatchInBackground = resultsStruct.BackgroundExecution; end end
Ren Okumura
Ren Okumura on 2 Nov 2022
同じ方法で行わせていただいたところ、私も元のイメージデータストアとそのラベル情報まで辿り着くことができました！
ご丁寧に教えていただきありがとうございました。大変お世話になりました！

Sign in to comment.

More Answers (0)

Categories

Find more on Convert Image Type in Help Center and File Exchange

Tags

Products


Release

R2022a

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!