MATLAB: Data augmentation in CNN

Deep Learning Toolboxhttps://in.mathworks.com/help/deeplearning/ref/trainnetwork.html#bu6sn60-2Image Processing Toolbox

Heloo… Iam working on a data set of 300 images of 2 classes. I want to perform the data augmentation. Below is the code i have attached for data augmentation. According to the methodology the accuracy must increase, but in our case it is decreasing from 96% to 87%. could anyone please help me in solving thia problem.
Thank you in advance:)
clc;
clear all;
close all;
myTrainingFolder = 'C:\Users\Admin\Desktop\Major Project\cnn_dataset';
%testingFolder = 'C:\Users\Be Happy\Documents\MATLAB\gtsrbtest';
imds = imageDatastore(myTrainingFolder,'IncludeSubfolders', true,  'LabelSource', 'foldernames');
%testingSet = imageDatastore(testingFolder,'IncludeSubfolders', true,  'LabelSource', 'foldernames');
labelCount = countEachLabel(imds);
numClasses = height(labelCount);
numImagesTraining = numel(imds.Files);
%% Create training and validation sets
[imdsTrainingSet, imdsValidationSet] = splitEachLabel(imds, 0.7, 'randomize');
%% Build a simple CNN 
imageSize = [227 227 3];
% Specify the convolutional neural network architecture.
layers = [
    imageInputLayer(imageSize)
    
    convolution2dLayer(3,8,'Padding','same')
    batchNormalizationLayer
    reluLayer   
    
    maxPooling2dLayer(2,'Stride',2)
    
    convolution2dLayer(3,16,'Padding','same')
    batchNormalizationLayer
    reluLayer   
    
    maxPooling2dLayer(2,'Stride',2)
    
    convolution2dLayer(3,32,'Padding','same')
    batchNormalizationLayer
    reluLayer   
    
    fullyConnectedLayer(numClasses)
    softmaxLayer
    classificationLayer];
%% Specify training options 
options = trainingOptions('sgdm', ...
    'MiniBatchSize',10, ...
    'MaxEpochs',6, ...
    'InitialLearnRate',1e-4, ...
    'Shuffle','every-epoch', ...
    'ValidationData',imdsValidationSet, ...
    'ValidationFrequency',3, ...
    'Verbose',false, ...
    'Plots','training-progress');
%% Train the network
net1 = trainNetwork(imdsTrainingSet,layers,options);
analyzeNetwork(net1);
%% Report accuracy of baseline classifier on validation set
YPred = classify(net1,imdsValidationSet);
YValidation = imdsValidationSet.Labels;
imdsAccuracy = sum(YPred == YValidation)/numel(YValidation);
%% Plot confusion matrix

figure, plotconfusion(YValidation,YPred)
%% PART 2: Baseline Classifier with Data Augmentation
%% Create augmented image data store
% Specify data augmentation options and values/ranges
imageAugmenter = imageDataAugmenter( ...
    'RandRotation',[-20,20], ...
    'RandXTranslation',[-5 5], ...
    'RandYTranslation',[-5 5]);
% Apply transformations (using randomly picked values) and build augmented
% data store
augImds = augmentedImageDatastore(imageSize,imdsTrainingSet, ...
    'DataAugmentation',imageAugmenter);
% (OPTIONAL) Preview augmentation results 
batchedData = preview(augImds);
figure, imshow(imtile(batchedData.input))
    
%% Train the network. 
net2 = trainNetwork(augImds,layers,options);
%% Report accuracy of baseline classifier with image data augmentation
YPred = classify(net2,imdsValidationSet);
YValidation = imdsValidationSet.Labels;
augImdsAccuracy = sum(YPred == YValidation)/numel(YValidation);
%% Plot confusion matrix
figure, plotconfusion(YValidation,YPred)

Best Answer

The following are few suggestions:
  1. Make sure that all the classes have equal number of observations.
  2. Check how trainNetwork uses an augmented image datastore to transform training data for each epoch: Augment Images for Training with Random Geometric Transformations. Then try training the network with augmentedImageDatastore for more epochs.
  3. Try changing the network architecture itself if there is no improvement in the accuracy when augmentedImageDatastore is used. You can refer to Choose Network Architecture.
  4. Try Using dropout layers & increasing global L2 regularization factor in new architecture. For more information, see dropoutLayer & 'L2Regularization' option in trainingOptions.
Related Question