OpenCV 4 with Python Blueprints - Second Edition

Training the model

Before we proceed with the actual training, it is a good idea to have some means to save the model with the best weights. For that purpose, we will use a callback from Keras:

checkpoint = K.callbacks.ModelCheckpoint("localization.h5",    monitor='val_root_mean_squared_error',    save_best_only=True, verbose=1)

The callback will be called after each epoch of training; it will calculate the root_mean_square_error metric of predictions on the validation data and will save the model to localization.h5 if the metric has improved.

Now, we train our model in the same way that we did with classification:

model.fit(    train.batch(32),    epochs=12,    validation_data=valid.batch(1),    callbacks=[checkpoint])

Here, the difference is that we train ...