Adding Source Code for Parallel Forall post on MATLAB Deep Learning

MATLAB_deeplearning/PetDetectionRecognitionScript.m

@@ -0,0 +1,111 @@
+% Copyright (c) 2015, MathWorks, Inc.
+
+%% Download and and predict using a pretrained ImageNet model
+
+% Download from MatConvNet pretrained networks repository
+urlwrite('http://www.vlfeat.org/matconvnet/models/imagenet-vgg-f.mat', 'imagenet-vgg-f.mat');
+cnnModel.net = load('imagenet-vgg-f.mat');
+
+% Setup up MatConvNet, modify the path if it's installed in a different
+% folder
+run(fullfile('matconvnet-1.0-beta15','matlab','vl_setupnn.m'));
+
+% Load and display an example image
+imshow('dog_example.png');
+img = imread('dog_example.png');
+
+% Predict label using ImageNet trained vgg-f CNN model
+label = cnnPredict(cnnModel,img);
+title(label,'FontSize',20)
+
+%% Load images from folder
+
+% Use imageSet to manage images stored in multiple folders
+imset = imageSet('pet_images','recursive');
+
+% Preallocate arrays with fixed size for prediction
+imageSize = cnnModel.net.normalization.imageSize;
+trainingImages = zeros([imageSize sum([imset(:).Count])],'single');
+
+% Load and resize images for prediction
+for ii = 1:numel(imset)
+    for jj = 1:imset(ii).Count
+        trainingImages(:,:,:,jj) = imresize(single(read(imset(ii),jj)),imageSize(1:2));
+    end
+end
+
+% Get the image labels directly from the ImageSet object
+trainingLabels = getImageLabels(imset);
+summary(trainingLabels)
+
+%% Extract features using pretrained CNN
+
+% Depending on how much memory you have on your GPU you may use a larger
+% batch size. We have 400 images, I'm going to choose 200 as my batch size
+cnnModel.info.opts.batchSize = 200;
+
+% Make prediction on a CPU
+[~, cnnFeatures, timeCPU] = cnnPredict(cnnModel,trainingImages,'UseGPU',false);
+% Make prediction on a GPU
+[~, cnnFeatures, timeGPU] = cnnPredict(cnnModel,trainingImages,'UseGPU',true);
+
+% Compare the performance increase
+bar([sum(timeCPU),sum(timeGPU)],0.5)
+title(sprintf('Approximate speedup: %2.00f x ',sum(timeCPU)/sum(timeGPU)))
+set(gca,'XTickLabel',{'CPU','GPU'},'FontSize',18)
+ylabel('Time(sec)'), grid on, grid minor
+
+%% Train a classifier using extracted features and calculate CV accuracy
+
+% Train and validate a linear support vector machine (SVM) classifier.
+classifierModel = fitcsvm(cnnFeatures, trainingLabels);
+
+% 10 fold crossvalidation accuracy
+cvmdl = crossval(classifierModel,'KFold',10);
+fprintf('kFold CV accuracy: %2.2f\n',1-cvmdl.kfoldLoss)
+
+%% Object Detection
+% Use findPet function that was automatically generated using the 
+% Image Region Analyzer App
+
+%% Tying the workflow together
+frameNumber = 0;
+vr = VideoReader(fullfile('PetVideos','videoExample.mov'));
+vw = VideoWriter('test.avi','Motion JPEG AVI');
+opticFlow = opticalFlowFarneback;
+open(vw);
+while hasFrame(vr)
+    % Count frames
+    frameNumber = frameNumber + 1;
+
+    % Step 1. Read Frame
+	videoFrame = readFrame(vr);
+
+    % Step 2. Detect ROI
+    vFrame = imresize(videoFrame,0.25);      % Get video frame
+    frameGray = rgb2gray(vFrame);            % Convert to gray for detection
+    bboxes = findPet(frameGray,opticFlow);   % Find bounding boxes
+    if ~isempty(bboxes)
+        img = zeros([imageSize size(bboxes,1)]);
+        for ii = 1:size(bboxes,1)
+            img(:,:,:,ii) = imresize(imcrop(vFrame,bboxes(ii,:)),imageSize(1:2));
+        end
+
+    % Step 3. Recognize object
+        % (a) Extract features using a CNN
+        [~, scores] = cnnPredict(cnnModel,img,'UseGPU',true,'display',false);
+
+        % (b) Predict using a trained Classifier
+        label = predict(classifierModel,scores);
+
+        % Step 4. Annotate object
+        vFrame = insertObjectAnnotation(vFrame,'Rectangle',bboxes,cellstr(label),'FontSize',40);
+    end
+
+    % Step 5. Write video to file
+      writeVideo(vw,videoFrame);
+
+%     fprintf('Frame: %d of %d\n',frameNumber,ceil(vr.FrameRate*vr.Duration));
+end
+close(vw);
+

MATLAB_deeplearning/README.md

@@ -0,0 +1,9 @@
+Companion Code for "Deep Learning for Computer Vision with MATLAB" by Shashank Prasanna
+============================
+
+This folder contains source code from the NVIDIA Parallel Forall Blog post [Deep Learning for Computer Vision with MATLABMATLAB by Joss Knight]() by Shashank Prasanna (The Mathworks).
+
+License
+-------
+
+These examples are released under the BSD open source license.  Refer to license.txt in this directory for full details.

MATLAB_deeplearning/cnnPredict.m

@@ -0,0 +1,88 @@
+function [classLabel, scores, batchTime] = cnnPredict(cnnModel,predImage,varargin)
+% Copyright (c) 2015, MathWorks, Inc.
+
+% Parse inputs
+p = inputParser;
+addParameter(p,'outputLayer',numel(cnnModel.net.layers),@isnumeric);
+addParameter(p,'UseGPU',false,@islogical);
+addParameter(p,'display',true,@islogical);
+parse(p,varargin{:});
+
+% Get batch size and number of images
+if ~isfield(cnnModel,'info')
+    cnnModel.info.opts.batchSize = 1;
+end
+batchSize = cnnModel.info.opts.batchSize;
+n_obs = size(predImage,4);
+isTapLayer = p.Results.outputLayer < numel(cnnModel.net.layers);
+
+if isTapLayer
+    cnnModel.net.layers(p.Results.outputLayer+1:end) = [];
+else
+    cnnModel.net.layers{end} = struct('type', 'softmax');
+end
+
+% Preallocate scores
+resTemp = vl_simplenn(cnnModel.net, cnnPreprocess(predImage(:,:,:,1)), [], []);
+scores = zeros([size(resTemp(end).x), n_obs]);
+
+% Move model to GPU if requested
+if p.Results.UseGPU
+    cnnModel.net = vl_simplenn_move(cnnModel.net,'gpu');
+end
+
+% Make predictions
+batchNumber = 0;
+numBatches = ceil(n_obs/batchSize);
+batchTime = zeros(numBatches,1);
+if p.Results.display
+    disp(' ')
+    fprintf('Using GPU: %s\n',mat2str(p.Results.UseGPU))
+    fprintf('Number of images: %d\n',n_obs)
+    fprintf('Number of batches: %d\n',numBatches)
+    fprintf('Number of layers in the Network: %d\n',numel(cnnModel.net.layers))
+    disp('-------------------------------------')
+end
+for ii = 1:batchSize:n_obs
+    tic
+    idx = ii:min(ii+batchSize-1,n_obs);
+	batchImages = predImage(:,:,:,idx);
+    im = cnnPreprocess(batchImages);
+
+	% Move batch to GPU if requested
+    if p.Results.UseGPU
+        im = gpuArray(im);
+    end
+    train_res = vl_simplenn(cnnModel.net, im, [], []);
+    scores(:,:,:,idx) = squeeze(gather(train_res(end).x));
+    batchNumber = batchNumber + 1;
+    batchTime(batchNumber) = toc;
+    if p.Results.display
+        fprintf('Batch: %2d/%d. Execution time: %2.4f\n',batchNumber,numBatches,batchTime(batchNumber))
+    end
+end
+
+if p.Results.display
+	fprintf('Avg. execution time/batch:   %2.4f\n',mean(batchTime))
+    disp('-------------------------------------')
+    fprintf('Total execution time:        %2.4f\n',sum(batchTime))
+    disp('-------------------------------------')
+end
+
+if isTapLayer
+    classLabel = [];
+else
+    scores = squeeze(gather(scores))';
+    [~, labelId] = max(scores,[],2);
+%     classLabel = categorical(cnnModel.net.classes.description(labelId)');
+    classLabel = cnnModel.net.classes.description(labelId)';
+end
+
+function im = cnnPreprocess(batchImages)
+    % Preprocess images
+    im = single(batchImages);
+    im = imresize(im, cnnModel.net.normalization.imageSize(1:2));
+	im = bsxfun(@minus,im,cnnModel.net.normalization.averageImage);
+end
+
+end

MATLAB_deeplearning/filterRegions.m

@@ -0,0 +1,22 @@
+function [BW_out,properties] = filterRegions(BW_in)
+% Copyright (c) 2015, MathWorks, Inc.
+%filterRegions  Filter BW image using auto-generated code from imageRegionAnalyzer app.
+%  [BW_OUT,PROPERTIES] = filterRegions(BW_IN) filters binary image BW_IN
+%  using auto-generated code from the imageRegionAnalyzer App. BW_OUT has
+%  had all of the options and filtering selections that were specified in
+%  imageRegionAnalyzer applied to it. The PROPERTIES structure contains the
+%  attributes of BW_out that were visible in the App.
+
+% Auto-generated by imageRegionAnalyzer app on 19-Oct-2015
+%---------------------------------------------------------
+
+BW_out = BW_in;
+
+% Filter image based on image properties.
+BW_out = bwpropfilt(BW_out, 'Area', [5000 + eps(5000), Inf]);
+
+% Get properties.
+properties = regionprops(BW_out, {'BoundingBox','Area'});
+
+% Uncomment the following line to return the properties in a table.
+% properties = struct2table(properties);

MATLAB_deeplearning/findPet.m

@@ -0,0 +1,12 @@
+function bboxes = findPet(frameGray, opticFlow)
+% Copyright (c) 2015, MathWorks, Inc.
+
+flow = estimateFlow(opticFlow,frameGray);
+threshImage = ( flow.Magnitude > 4);
+[BW_out,regions] = filterRegions(threshImage);
+if(size(regions) > 0)
+    bboxes = regions.BoundingBox;
+else
+    bboxes = [];
+end
+end

MATLAB_deeplearning/getImageLabels.m

@@ -0,0 +1,5 @@
+function imageType = getImageLabels(imset)
+% Copyright (c) 2015, MathWorks, Inc.
+    imageType = categorical(repelem({imset.Description}', ...
+        [imset.Count], 1));
+end