Convolutional Neural Network for traffic signs classification.

Hello,

I want to create traffic sign classifier with CNN based on dataset from GTSRB. I made my net only for 11 classes. When I try to classify picture of sign which is in my dataset then I have almost everytime 100% accuracy, but when I try to cfassily sign which isn't in my dataset ( for example speed limit 20 km/h image downloaded from google ) the prediction is mostly incorrect.

How can I improve my network ?

Dataset : https://sid.erda.dk/public/archives/daaeac0d7ce1152aea9b61d9f1e19370/published-archive.html

Code :

clc;clear;
DatasetPath = fullfile('C:\Users\Pulpit\Splotowe Sieci Neuronowe\GTSRB\fl');
Data = imageDatastore(DatasetPath, ...
    'IncludeSubfolders',true,'LabelSource','foldernames');
[imdsTrain,imdsValidation] = splitEachLabel(Data,0.7,'randomize');
imageAugmenter = imageDataAugmenter( ...
    'RandRotation',[-20,20], ...
    'RandXTranslation',[-1 1], ...
    'RandYTranslation',[-1 1], ...
    'RandXReflection', false, ...
    'RandYReflection', true ...
    );

imageSize = [32 32 3];
augimdsTrain = augmentedImageDatastore(imageSize,imdsTrain,'DataAugmentation',imageAugmenter);
augimdsValidation = augmentedImageDatastore(imageSize(1:2),imdsValidation);

%%
layers = [
    imageInputLayer([32 32 3])
    
    convolution2dLayer(3,64,'Stride' ,1,'Padding','same' )
    convolution2dLayer(3,64,'Stride' ,1,'Padding','same' )

    batchNormalizationLayer
    reluLayer   
    maxPooling2dLayer(1, 'Stride', 1);
 
    convolution2dLayer(3,64,'Stride' ,1,'Padding','same' )
    convolution2dLayer(3,64,'Stride' ,1,'Padding','same' )
    batchNormalizationLayer
    reluLayer   
    maxPooling2dLayer(1, 'Stride', 1);
    convolution2dLayer(3,128,'Stride' ,1,'Padding','same' )
    convolution2dLayer(3,128,'Stride' ,1,'Padding','same' )
    convolution2dLayer(3,128,'Stride' ,1,'Padding','same' )
    batchNormalizationLayer
    reluLayer   
    maxPooling2dLayer(1, 'Stride', 1);
    convolution2dLayer(3,256,'Stride' ,1,'Padding','same' )
    convolution2dLayer(3,256,'Stride' ,1,'Padding','same' )
    convolution2dLayer(3,256,'Stride' ,1,'Padding','same' )
    batchNormalizationLayer
    reluLayer   
    maxPooling2dLayer(1, 'Stride', 1);
    fullyConnectedLayer(64)
    fullyConnectedLayer(32)
    fullyConnectedLayer(11)
    softmaxLayer
    classificationLayer];
%miniBatchSize = 32;
%valFrequency = floor(numel(imdsValidation.Files)/miniBatchSize);
%%
%{
options = trainingOptions('sgdm', ...
'LearnRateSchedule','piecewise', ...
    'LearnRateDropFactor',0.2, ...
    'LearnRateDropPeriod',5, ...   
        'ValidationData',augimdsValidation, ...
        'ValidationFrequency',valFrequency, ...
'InitialLearnRate',0.001, ...
    'MaxEpochs',10, ...
    'MiniBatchSize',miniBatchSize, ...
    'Plots','training-progress');
%}
options = trainingOptions('sgdm', ...
    'MiniBatchSize',10, ...
    'MaxEpochs',6, ...
    'InitialLearnRate',1e-4, ...
    'Shuffle','every-epoch', ...
    'ValidationData',augimdsValidation, ...
    'ValidationFrequency',3, ...
    'Verbose',false, ...
    'Plots','training-progress');
%%
convnet = trainNetwork(augimdsTrain,layers,options);
%%
YTest = imdsValidation.Labels;
[YPred,prob] = classify(convnet,augimdsValidation);
accuracy = sum(YPred == YTest)/numel(YTest)
plotconfusion(YTest,YPred)


idx = randperm(numel(imdsValidation.Files),4);
figure
for i = 1:4
    subplot(2,2,i)
    I = readimage(imdsValidation,idx(i));
    imshow(I)
    label = YPred(idx(i));
    title(string(label) + ", " + num2str(100*max(prob(idx(i),:)),3) + "%");
end

[Omran Adnanoglu]

Hi Krystian,

I am working on the same project, i still in the very beginning and i am facing a problem while reading the data with the given code like this :

Error using imread (line 438)

End of file reached too early.

Error in TrainTrafficSigns (line 28)

Img = imread(ImgFile);

Can you please share me the part of code where you read the data ?

[Krystian P]

Hi,

I did it in this way:

DatasetPath = fullfile('Your Path');

Data = imageDatastore(DatasetPath, ...

'IncludeSubfolders',true,'LabelSource','foldernames');

But this is for folder with subfolders, where every subfolder is different class of road sign.

How do you want to do this ?

If you want read single image maybe try this:

Img = imread('Filename'); or Img = imread(FilePath);

Best regards.

Answer 1

[ZHI-RUI LIN]

Excuse me, i would like to ask how you loaded the GTSRB data set (ppm file) into matlab, thank you !