Hi Tai,
I understand that you are looking for a way to increase the number of epochs for training a model which is already trained for certain epochs.
To continue training from where you left off, you can use the previously trained detector as the starting point for further training. When you train for the first time and get the `detector` object, you can use this object as an input to `trainYOLOv2ObjectDetector` for subsequent training sessions.
Here's what you need to do:
1. After your first training session, save the trained `detector` object to a file:
[detector, info] = trainYOLOv2ObjectDetector(dsTrain, lgraph, opts);
save('trainedDetector.mat', 'detector');
2. When you want to continue training, load the `detector` object from the file and use it as the network input to `trainYOLOv2ObjectDetector`. Also, adjust the `MaxEpochs` in your `trainingOptions` to account for the additional epochs you want to train for:
% Load the previously trained detector
load('trainedDetector.mat', 'detector');
% Adjust the training options for additional epochs
opts = trainingOptions("rmsprop",...
InitialLearnRate=0.001,...
MiniBatchSize=4,...
MaxEpochs=10, ... % Total epochs you want to train (including previous epochs)
LearnRateSchedule="piecewise",...
LearnRateDropPeriod=5,...
VerboseFrequency=30, ...
L2Regularization=0.001,...
ValidationData=dsVal, ...
ValidationFrequency=50, ...
OutputNetwork="best-validation-loss", ...
Plots="training-progress");
% Continue training
[detector, info] = trainYOLOv2ObjectDetector(dsTrain, detector.Network, opts);
In the code above, `MaxEpochs` is set to 10, which assumes that you want to train for a total of 10 epochs, including the 5 epochs you've already completed. The training will resume from the state saved in `detector.Network`.
Make sure to replace `'trainedDetector.mat'` with the actual path to your saved detector file.
By using the saved `detector` object, you're effectively continuing the training from the last saved state, which includes the learned weights and biases up to that point. This way, you don't have to train from scratch, and you can incrementally improve your model by training for additional epochs as needed.
Hope it helps!
