Data Augmentation Image Data Generator Keras Semantic Segmentation

Question

I'm fitting full convolutional network on some image data for semantic segmentation using Keras. However, I'm having some problems overfitting. I don't have that much data and I want to do data augmentation. However, as I want to do pixel-wise classification, I need any augmentations like flips, rotations, and shifts to apply to both feature images and the label images. Ideally I'd like to use the Keras ImageDataGenerator for on-the-fly transformations. However, as far as I can tell, you cannot do equivalent transformations on both the feature and label data.

Does anyone know if this is the case and if not, does anyone have any ideas? Otherwise, I'll use other tools to create a larger dataset and just feed it in all at once.

Thanks!

Gott

I had exactly the same problem, I also thought to ImageDataGenerator and I didn't find any solution. So I did it myself :-( – Inessential

Could I ask what tool you ended up using for this process to ensure that the same transformations were applied to both the features and the data? – Gott

You can do this relatively easily by creating your own batch generator where you augment inputs/outputs the same way and then call model.train_on_batch. Make sure to shuffle the data yourself as this is normally taken care of by model.fit. – Shipment

My own java code that I apply to the batch, or the batch + ground truth. – Inessential

Answer 1

There are works on extending ImageDataGenerator to be more flexible for exactly these type of cases (see in this issue on Github for examples).

Additionally, as mentioned by Mikael Rousson in the comments, you can easily create your own version of ImageDataGenerator yourself, while leveraging many of its built-in functions to make it easier. Here is an example code I've used for an image denoising problem, where I use random crops + additive noise to generate clean and noisy image pairs on the fly. You could easily modify this to add other types of augmentations. After which, you can use Model.fit_generator to train using these methods.

from keras.preprocessing.image import load_img, img_to_array, list_pictures

def random_crop(image, crop_size):
    height, width = image.shape[1:]
    dy, dx = crop_size
    if width < dx or height < dy:
        return None
    x = np.random.randint(0, width - dx + 1)
    y = np.random.randint(0, height - dy + 1)
    return image[:, y:(y+dy), x:(x+dx)]

def image_generator(list_of_files, crop_size, to_grayscale=True, scale=1, shift=0):
    while True:
        filename = np.random.choice(list_of_files)
        try:
            img = img_to_array(load_img(filename, to_grayscale))
        except:
            return
        cropped_img = random_crop(img, crop_size)
        if cropped_img is None:
            continue
        yield scale * cropped_img - shift
def corrupted_training_pair(images, sigma):
    for img in images:
        target = img
        if sigma > 0:
            source = img + np.random.normal(0, sigma, img.shape)/255.0
        else:
            source = img
        yield (source, target)
def group_by_batch(dataset, batch_size):
    while True:
        try:
            sources, targets = zip(*[next(dataset) for i in xrange(batch_size)])
            batch = (np.stack(sources), np.stack(targets))
            yield batch
        except:
            return
def load_dataset(directory, crop_size, sigma, batch_size):
    files = list_pictures(directory)
    generator = image_generator(files, crop_size, scale=1/255.0, shift=0.5)
    generator = corrupted_training_pair(generator, sigma)
    generator = group_by_batch(generator, batch_size)
    return generator

You can then use the above like so:

train_set = load_dataset('images/train', (patch_height, patch_width), noise_sigma, batch_size)
val_set = load_dataset('images/val', (patch_height, patch_width), noise_sigma, batch_size)
model.fit_generator(train_set, samples_per_epoch=batch_size * 1000, nb_epoch=nb_epoch, validation_data=val_set, nb_val_samples=1000)

Meaningful

What does the flow function do while using data generators in Keras ? – Bernadinebernadotte

The flow function is for creating a data generator from directory of files or numpy arrays. However, this have nothing with this question, or my answer. I suggest you open a new question with your exact problem and not as a comment on an answer to an unrelated question. – Meaningful

Answer 2

Yes you can. Here's an example from Keras's docs. You zip together two generators seeded with the same seeds and the fit_generator them. https://keras.io/preprocessing/image/

# we create two instances with the same arguments 
data_gen_args = dict(featurewise_center=True,
                     featurewise_std_normalization=True,
                     rotation_range=90.,
                     width_shift_range=0.1,
                     height_shift_range=0.1,
                     zoom_range=0.2) 
image_datagen = ImageDataGenerator(**data_gen_args) 
mask_datagen = ImageDataGenerator(**data_gen_args)

# Provide the same seed and keyword arguments to the fit and flow methods seed = 1 
image_datagen.fit(images, augment=True, seed=seed) 
mask_datagen.fit(masks, augment=True, seed=seed)

image_generator = image_datagen.flow_from_directory(
    'data/images',
    class_mode=None,
    seed=seed)

mask_generator = mask_datagen.flow_from_directory(
    'data/masks',
    class_mode=None,
    seed=seed)

# combine generators into one which yields image and masks 
train_generator = zip(image_generator, mask_generator)

model.fit_generator(
    train_generator,
    samples_per_epoch=2000,
    nb_epoch=50)

Agog

i get stuck when i try to zip them together, do you have an idea? train_generator = zip(image_generator, mask_generator)either memory limit or never ending execution – Mozambique

use itertools.izip() instead, that gives you the generator version of zip() – Leapfrog

@Dennis I've a onehot mask encoding function and would like to convert all the masks to this encoding. How/where I can call my function in terms of the mask_generator? – Ohaus

flow_from_directory() requires subdirectory represent each class. Only when your data format in this way . Or you have to format before using it. – Huntingdonshire

I can't figure out what images and masks should be. I know they represent input data X and target data Y but I thought X and Y are getting pulled from the generators.. Hmm. – Conciliator

I think this is exclusively for training datasets. You still need to include the data for validation, which will demand more two blocks of flow_from_directory(...). I've tried ImageDataGenerator(..., validation_split=0.15, ...) (see here) among the params, but got 99% of accuracy. I don't know, did not feel good to use validation_split, then, splited the data by myself and added two more blocks as mentioned. – Camper