Trjtdtk

I am using flow_from_directory() and fit_generator in my deep learning model, and I want to use cross validation method to train the CNN model.

datagen = ImageDataGenerator(rotation_range=15,width_shift_range=0.2,
 height_shift_range=0.2,shear_range=0.2,
 zoom_range=0.2,horizontal_flip=True,
 fill_mode='nearest')

image_size = (224, 224)
batch = 32

train_generator = datagen.flow_from_directory(train_data,
 target_size=image_size,
 batch_size=batch,
 classes= classes_array)

I found this Youtube video and this Tutorial, But it is not use flow_from_directory().

Do you have any idea how do I use k-fold cross validation when using fit_generator and flow_from_directory() in Keras?

So, I haven't found any solution regarding this application of cross-validation in fit_generator(), I hope it comes in one update of the Keras package, since cross-validation is an important part of training models.

What I have done so far, basically I split the dataset first then I pass the data and labels to the fit_generator. At the end of each step I save the model, at the beggining of each step I load the preciuos model to continue the training over the same model, but with a different k-fold of the dataset. I'm pasting the part of my code where I use this approach, hope it helps.

## Training with K-fold cross validation
kf = KFold(n_splits=k_folds, random_state=None, shuffle=True)
kf.get_n_splits(images_file_paths)

X = np.array(images_file_paths)
y = np.array(class_labels)

i = 1
for train_index, test_index in kf.split(X):
 trainData = X[train_index]
 testData = X[test_index]
 trainLabels = y[train_index]
 testLabels = y[test_index]

 print("=========================================")
 print("====== K Fold Validation step => %d/%d =======" % (i,k_folds))
 print("=========================================")

 trainGenerator = Generator(trainData,trainLabels,batchSize=batchSize,imageSize=imageSize,augment=True,grayMode=grayMode)
 valGenerator = Generator(testData,testLabels,batchSize=batchSize,imageSize=imageSize,augment=False,grayMode=grayMode)

 try: 
 model = load_model(weights_path, compile=True)

 except Exception as OSError:
 pass

 model.fit_generator(
 trainGenerator,
 steps_per_epoch=len(trainData),
 epochs=epochs,
 validation_data=valGenerator,
 validation_steps=len(testData))
 i+=1