load('object-detection-crowdai_training_label.mat');

training_index = 1:4000;

%training_index = 1:20;

crowdai_training_label.imageFilename = replace(crowdai_training_label.imageFilename,'\','/');

training_data = crowdai_training_label(training_index,:);

inputLayer = imageInputLayer([1200 1920 3]);

middleLayers = [

    convolution2dLayer([11 11], 96,'stride',4, 'Padding', 0, 'NumChannels',3 )

    reluLayer()

    crossChannelNormalizationLayer(5)

    maxPooling2dLayer(3,'stride',2, 'Padding',0)

    convolution2dLayer([3 3], 384,'stride',1, 'Padding', 1,'NumChannels',96)

    reluLayer()

    convolution2dLayer([3 3], 256,'stride',1, 'Padding', 1,'NumChannels',384)

    reluLayer()

    maxPooling2dLayer(3,'stride',2, 'Padding',0)

    ];

finalLayers = [

    fullyConnectedLayer(500)

    reluLayer()

    dropoutLayer(0.5)

    fullyConnectedLayer(100)

    reluLayer()

    dropoutLayer(0.5)

    fullyConnectedLayer(width(training_data))

      % Add the softmax loss layer and classification layer. 

      softmaxLayer()

      classificationLayer()

  ];

layers = [

    inputLayer

    middleLayers

    finalLayers

    ]

optionsStage1 = trainingOptions('sgdm', ...

    'MiniBatchSize', 50, ...

    'MaxEpochs', 5, ...

    'InitialLearnRate', 1e-3, ...

    'CheckpointPath', tempdir);

% Options for step 2

optionsStage2 = trainingOptions('sgdm', ...

    'MiniBatchSize', 50, ...

    'MaxEpochs', 5, ...

    'InitialLearnRate', 1e-3, ...

    'CheckpointPath', tempdir);

% Options for step 3.

optionsStage3 = trainingOptions('sgdm', ...

    'MiniBatchSize', 50, ...

    'MaxEpochs', 5, ...

    'InitialLearnRate', 1e-3, ...

    'CheckpointPath', tempdir);

% Options for step 4.

optionsStage4 = trainingOptions('sgdm', ...

    'MiniBatchSize', 50, ...

    'MaxEpochs', 5, ...

    'InitialLearnRate', 1e-3, ...

    'CheckpointPath', tempdir);

options = [

    optionsStage1

    optionsStage2

    optionsStage3

    optionsStage4

    ];

%%

tic

% A trained network is loaded from disk to save time when running the

% example. Set this flag to true to train the network. 

% Set random seed to ensure example training reproducibility.

rng(0);

  % Train Faster R-CNN detector. Select a BoxPyramidScale of 1.2 to allow

  % for finer resolution for multiscale object detection.

detector = trainFasterRCNNObjectDetector(training_data, layers, options, ...

  'NegativeOverlapRange', [0 0.3], ...

  'PositiveOverlapRange', [0.7 1], ...

  'SmallestImageDimension', 1920, ...

  'BoxPyramidScale', 1.2);

toc

save('crowdai_fasterRCNN_detector.mat','detector');

%%

testData = crowdai_training_label;

%% Read a test image.

I = imread(testData.imageFilename{3756});

% figure(1);

% imshow(I);

% Run the detector.

[bboxes,scores] = detect(detector,I);

%% Annotate detections in the image.

I = insertObjectAnnotation(I,'rectangle',bboxes,scores);

savefig(I);

%%

test_index = 5000:9000;

test_data = crowdai_training_label(test_index,:);

%%

disp('Processing...detector')

tic

% Run detector on each image in the test set and collect results.

resultsStruct = struct([]);

for i = 1:length(test_index)

    % Read the image.

    I = imread(test_data.imageFilename{i});

    % Run the detector.

    [bboxes, scores, labels] = detect(detector, I);

    % Collect the results.

    resultsStruct(i).Boxes = bboxes;

    resultsStruct(i).Scores = scores;

    resultsStruct(i).Labels = labels;

end

% Convert the results into a table.

results = struct2table(resultsStruct);

% Extract expected bounding box locations from test data.

expectedResults = test_data(:, 2:end);

% Evaluate the object detector using Average Precision metric.

[ap, recall, precision] = evaluateDetectionPrecision(results, expectedResults);

toc

%

% Plot precision/recall curve

h = figure;

plot(recall,precision)

xlabel('Recall')

ylabel('Precision')

grid on

title(sprintf('Average Precision = %.2f', ap))

save(h,'PR curve','jpg');

disp('end');