Why cannot the Faster-RCNN run but Fast-RCNN can with the same object detection data and based on the samples from Matlab?

Hi there,

Recently, I have being learned how to use deep learning to detect object. I am confused a lot with a question about Faster-RCNN.

Why cannot the Faster-RCNN run but Fast-RCNN can with the same object detection data and based on the samples from Matlab?

Below are the part of codes for the Faster-RCNN rand Fast-RCNN.

%% Train Faster R-CNN building Detector

idx = floor(0.6 * height(wlDataset));

trainingData = wlDataset(1:idx,:);

testData = wlDataset(idx:end,:);

inputLayer = imageInputLayer([256 256 3]);

filterSize = [3 3];

numFilters = 32;

% Create the middle layers.

middleLayers = [

convolution2dLayer(filterSize, numFilters, 'Padding', 1)

reluLayer()

convolution2dLayer(filterSize, numFilters, 'Padding', 1)

reluLayer()

maxPooling2dLayer(3, 'Stride',2)

];

finalLayers = [

% Add a ReLU non-linearity.

reluLayer()

fullyConnectedLayer(width(wlDataset))

% Add the softmax loss layer and classification layer.

softmaxLayer()

classificationLayer()

];

% Combine the input, middle, and final layers.

layers = [

inputLayer

middleLayers

finalLayers

];

optionsStage1 = trainingOptions('sgdm', ...

'MaxEpochs', 10, ...

'MiniBatchSize', 1, ...

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

'CheckpointPath', tempdir);

% Options for step 2.

optionsStage2 = trainingOptions('sgdm', ...

'MaxEpochs', 10, ...

'MiniBatchSize', 1, ...

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

'CheckpointPath', tempdir);

% Options for step 3.

optionsStage3 = trainingOptions('sgdm', ...

'MaxEpochs', 10, ...

'MiniBatchSize', 1, ...

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

'CheckpointPath', tempdir);

% Options for step 4.

optionsStage4 = trainingOptions('sgdm', ...

'MaxEpochs', 10, ...

'MiniBatchSize', 1, ...

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

'CheckpointPath', tempdir);

options = [

optionsStage1

optionsStage2

optionsStage3

optionsStage4

];

%%

doTrainingAndEval = true;

if doTrainingAndEval

% Set random seed to ensure example training reproducibility.

rng(1);

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

% for finer resolution for multiscale object detection.

fasterrcnn = trainFasterRCNNObjectDetector(trainingData, layers, options, ...

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

'PositiveOverlapRange', [0.6 1], ...

'NumRegionsToSample', [256 128 256 128], ...

'BoxPyramidScale', 1.2);

else

% Load pretrained detector for the example.

% detector = data.detector;

end

============================================

%% Train Fast R-CNN building Detector

clc;

wlDataset = wlData;

idx = floor(0.6 * height(wlDataset));

trainingData = wlDataset(1:idx,:);

testData = wlDataset(idx:end,:);

inputLayer = imageInputLayer([32 32 3]);

filterSize = [3 3];

numFilters = 32;

% Create the middle layers.

middleLayers = [

convolution2dLayer(filterSize, numFilters, 'Padding', 1)

reluLayer()

convolution2dLayer(filterSize, numFilters, 'Padding', 1)

reluLayer()

maxPooling2dLayer(3, 'Stride',2)

];

finalLayers = [

fullyConnectedLayer(64)

reluLayer()

fullyConnectedLayer(width(wlDataset))

softmaxLayer()

classificationLayer()

];

layers = [

inputLayer

middleLayers

finalLayers

];

options = trainingOptions('sgdm', ...

'MiniBatchSize', 1, ...

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

'MaxEpochs', 10, ...

'CheckpointPath', tempdir);

frcnn = trainFastRCNNObjectDetector(trainingData, layers, options);