Hello @Eric Louchard, in MATLAB, you can add a bounding box regression layer to a non-R-CNN network using the 'regressionLayer' function. The 'regressionLayer' function can be used to create a custom bounding box regression layer in MATLAB.
Here's an example of how to add a bounding box regression layer to the simple character recognition network in MATLAB:
% Load the example data
[XTrain, YTrain] = digitTrain4DArrayData;
% Define the network layers
layers = [
imageInputLayer([28 28 1])
convolution2dLayer(5, 20)
reluLayer
maxPooling2dLayer(2, 'Stride', 2)
fullyConnectedLayer(10)
softmaxLayer
classificationLayer
regressionLayer % Add the regression layer after the classification layer
];
% Define the training options
options = trainingOptions('sgdm', ...
'MaxEpochs', 20, ...
'MiniBatchSize', 128, ...
'Plots', 'training-progress');
% Train the network with the regression layer
net = trainNetwork(XTrain, YTrain, layers, options);
In this example, the 'regressionLayer' function is added after the 'classificationLayer' to create a custom bounding box regression layer. You can then train the network with the new layer using the 'trainNetwork' function.
Note that the 'regressionLayer' function in MATLAB assumes that the input to the layer is a 4-D array of size [height, width, channels, batchsize]. Therefore, you'll need to reshape your data accordingly before using the 'regressionLayer' function.
