def load_ciarp(dataset_name):
"""
Load ciarp dataset.
Returns (x, y): as dataset x and y.
"""
dataset_path = '/develop/data/{}/data'.format(dataset_name)
if not os.path.exists(dataset_path):
os.makedirs(dataset_path)
data = hd.load(dataset_name, dataset_path)
# TODO: define best way to do this
x_train, y_train = data['train_Kinect_WithoutGabor']
x_test, y_test = data['test_Kinect_WithoutGabor']
x = np.concatenate((x_train, x_test), axis=0)
y = np.concatenate((y_train, y_test), axis=0)
return x, y
def load_lsa16(dataset_name):
"""
Load lsa16 dataset.
Returns (x, y): as dataset x and y.
"""
DATASET_PATH = '/develop/data/{}/data'.format(dataset_name)
if not os.path.exists(DATASET_PATH):
os.makedirs(DATASET_PATH)
data = hd.load(dataset_name, DATASET_PATH)
# TODO: define best way to do this
x_train, y_train = data['train_Kinect_WithoutGabor']
x_test, y_test = data['test_Kinect_WithoutGabor']
X = np.concatenate((x_train, x_test), axis=0)
y = np.concatenate((y_train, y_test), axis=0)
return X, y
def load_rwth(data_dir, config, splits):
"""
Load rwth dataset.
Args:
data_dir (str): path of the directory with 'splits', 'data' subdirs.
config (dict): general dict with program settings.
splits (list): list of strings 'train'|'val'|'test'
Returns (dict): dictionary with keys as splits and values as tf.Dataset
"""
DATASET_NAME = "rwth"
DATASET_PATH = "/develop/data/rwth/data"
data = hd.load(DATASET_NAME, DATASET_PATH)
features = data[0]
classes = data[1]['y']
good_min = 20
good_classes = []
for i in range(len(classes)):
images = features[np.equal(i, classes)]
if len(images) >= good_min:
good_classes = good_classes + [i]
good_x = features[np.in1d(classes, good_classes)]
good_y = classes[np.in1d(classes, good_classes)]
my_dict = dict(zip(np.unique(good_y), range(len(np.unique(good_y)))))
good_y = np.vectorize(my_dict.get)(good_y)
features, classes = good_x, good_y
uniqueClasses = np.unique(classes)
x_train, x_test, y_train, y_test = train_test_split_balanced(
features,
classes,
train_size=config['data.train_size'],
test_size=config['data.test_size'])
x_train, x_test = x_train / 255.0, x_test / 255.0
_, amountPerTrain = np.unique(y_train, return_counts=True)
_, amountPerTest = np.unique(y_test, return_counts=True)
train_datagen_args = dict(
featurewise_center=True,
featurewise_std_normalization=True,
rotation_range=config['data.rotation_range'],
width_shift_range=config['data.width_shift_range'],
height_shift_range=config['data.height_shift_range'],
horizontal_flip=config['data.horizontal_flip'],
fill_mode='constant',
cval=0)
train_datagen = ImageDataGenerator(train_datagen_args)
train_datagen.fit(x_train)
test_datagen_args = dict(featurewise_center=True,
featurewise_std_normalization=True,
fill_mode='constant',
cval=0)
test_datagen = ImageDataGenerator(test_datagen_args)
test_datagen.fit(x_train)
w, h, c = list(map(int, config['model.x_dim'].split(',')))
ret = {}
for split in splits:
# n_way (number of classes per episode)
if split in ['val', 'test']:
n_way = config['data.test_way']
else:
n_way = config['data.train_way']
# n_support (number of support examples per class)
if split in ['val', 'test']:
n_support = config['data.test_support']
else:
n_support = config['data.train_support']
# n_query (number of query examples per class)
if split in ['val', 'test']:
n_query = config['data.test_query']
else:
n_query = config['data.train_query']
if split in ['val', 'test']:
y = y_test
x = x_test
dg = train_datagen
dg_args = train_datagen_args
else:
y = y_train
x = x_train
dg = test_datagen
dg_args = test_datagen_args
amountPerClass = amountPerTest if split in ['val', 'test'
] else amountPerTrain
i = np.argsort(y)
x = x[i, :, :, :]
for index in i:
x[index, :, :, :] = dg.apply_transform(x[index], dg_args)
data = np.reshape(x,
(len(uniqueClasses), amountPerClass[0], 132, 92, 3))
data_loader = DataLoader(data,
n_classes=len(uniqueClasses),
n_way=n_way,
n_support=n_support,
n_query=n_query,
x_dim=(w, h, c))
ret[split] = data_loader
return ret
def train(self, dataset_id, epochs, batch_size=128, save_interval=50):
# Load the dataset
#(X_train, _), (_, _) = mnist.load_data()
x, metadata = hd.load(dataset_id)
X_train, X_test, Y_train, Y_test = self.split(
parameters.get_split_value(dataset_id), x, metadata['y'])
# Rescale -1 to 1
X_train = (X_train.astype(np.float32) - 127.5) / 127.5
#X_train = np.expand_dims(X_train, axis=3)
#half_batch = int(batch_size / 2)
valid = np.ones((batch_size, 1))
fake = np.zeros((batch_size, 1))
for epoch in range(epochs):
# ---------------------
# Train Discriminator
# ---------------------
# Select a random half batch of images
idx = np.random.randint(0, X_train.shape[0], batch_size)
imgs, labels = X_train[idx], Y_train[idx]
noise = np.random.normal(0, 1, (batch_size, self.noise_value))
# Generate a half batch of new images
gen_imgs = self.generator.predict([noise, labels])
# Train the discriminator
d_loss_real = self.discriminator.train_on_batch([imgs, labels],
valid)
d_loss_fake = self.discriminator.train_on_batch([gen_imgs, labels],
fake)
d_loss = 0.5 * np.add(d_loss_real, d_loss_fake)
# ---------------------
# Train Generator
# ---------------------
sampled_labels = np.random.randint(0, self.classes,
batch_size).reshape(-1, 1)
# The generator wants the discriminator to label the generated samples
# as valid (ones)
#valid_y = np.array([1] * batch_size)
# Train the generator
g_loss = self.combined.train_on_batch([noise, sampled_labels],
valid)
# Plot the progress
print("%d [D loss: %f, acc.: %.2f%%] [G loss: %f]" %
(epoch, d_loss[0], 100 * d_loss[1], g_loss))
# If at save interval => save generated image samples
if epoch % save_interval == 0:
self.save_imgs(epoch)
save_path = os.path.join(default_folder, self.name)
self.base_model.save(
os.path.join(save_path,
f"{self.name}_GANdiscriminator{epochs}.h5"))
self.base_model.save_weights(
os.path.join(save_path,
f"{self.name}_GANdiscriminator{epochs}_weights.h5"))
print("Saved model to disk")
asl_class += 1
if (index == 62):
asl_class += 2
if (index == 86):
asl_class += 1
if (index == -1):
worksheet.write(0, col, f"{index}")
index += 1
else:
worksheet.write(0, col, f"{index}({asl_class})")
worksheet.set_column(col, 32)
index += 1
asl_class += 1
for i, dataset_id in enumerate(hd.ids()):
x, metadata = hd.load(dataset_id)
worksheet.write(i + 1, 0, dataset_id)
worksheet.set_row(i + 1, 32)
flag = np.zeros(metadata['y'].max() + 1)
if (x.shape[3] == 1):
x = np.repeat(x, 3, -1)
for h in range(len(x)):
clas = metadata['y'][h]
if (flag[clas] == 0):
path_to_save = cache_path / f"{dataset_id}image{h}.png"
if (dataset_id == "PugeaultASL_B"):
img_depth = np.zeros((x[h].shape[0], x[h].shape[1]),
dtype='f8')
max_z = x[h].max()
def __init__(self, dataset_id, **kwargs):
if 'version' in kwargs:
ver = kwargs['version']
if 'delete' in kwargs:
supr = kwargs['delete']
try:
self.dataset = hd.load(dataset_id, version=ver, delete=supr)
except:
try:
self.dataset = hd.load(dataset_id, version=ver)
except:
try:
self.dataset = hd.load(dataset_id, delete=supr)
except:
self.dataset = hd.load(dataset_id)
self.input_shape = self.dataset[0][0].shape
self.img_rows = (self.input_shape[0] // 4) * 4
self.img_cols = (self.input_shape[1] // 4) * 4
self.channels = 3
self.name = dataset_id
if (self.name == "psl" or self.name == "indianB"):
self.img_shape = (128, 128, self.channels)
self.img_rows = 128
self.img_cols = 128
else:
if (self.name == "indianA"):
self.img_shape = (64, 64, self.channels)
self.img_rows = 64
self.img_cols = 64
else:
self.img_shape = (self.img_rows, self.img_cols, self.channels)
self.classes = self.dataset[1]['y'].max() + 1
self.noise_value = 100
optimizer = Adam(0.0002, 0.5)
# Build and compile the discriminator
self.base_model, self.discriminator = self.build_discriminator()
self.discriminator.compile(loss='binary_crossentropy',
optimizer=optimizer,
metrics=['accuracy'])
# Build and compile the generator
self.generator = self.build_generator()
noise = keras.layers.Input(shape=(self.noise_value, ))
label = keras.layers.Input(shape=(1, ))
img = self.generator([noise, label])
#self.generator.compile(loss='binary_crossentropy', optimizer=optimizer)
# The generator takes noise as input and generated imgs
#z = keras.layers.Input(shape=(100,))
#img = self.generator(z)
# For the combined model we will only train the generator
self.discriminator.trainable = False
# The valid takes generated images as input and determines validity
valid = self.discriminator([img, label])
# The combined model (stacked generator and discriminator) takes
# noise as input => generates images => determines validity
self.combined = Model([noise, label], valid)
self.combined.compile(loss='binary_crossentropy', optimizer=optimizer)
self.path = default_folder
if not os.path.exists(self.path):
os.makedirs(self.path)
def load_rwth(config, path=None):
"""
Load rwth dataset.
Returns (x, y): as dataset x and y.
"""
train_size = config['data.train_size']
test_size = config['data.test_size']
n_train_per_class = config['data.n_train_per_class']
n_test_per_class = config['data.n_test_per_class']
if path == None:
path = '/tf/data/{}'.format(config['data.dataset'])
data_dir = os.path.join(path, 'data')
if not os.path.exists(data_dir):
os.makedirs(data_dir)
data = hd.load(config['data.dataset'], Path(data_dir))
if config['data.split']:
split_dir = os.path.join(path, 'splits', config['data.split'])
def split_file(split):
return os.path.join(split_dir, f"{split}.txt")
x_train, y_train = load_from_split(config['data.dataset'],
config['data.version'], data_dir,
split_file('train'))
x_test, y_test = load_from_split(config['data.dataset'],
config['data.version'], data_dir,
split_file('test'))
x_val, y_val = load_from_split(config['data.dataset'],
config['data.version'], data_dir,
split_file('val'))
else:
good_min = 40
good_classes = []
n_unique = len(np.unique(data[1]['y']))
for i in range(n_unique):
images = data[0][np.equal(i, data[1]['y'])]
if len(images) >= good_min:
good_classes = good_classes + [i]
x = data[0][np.in1d(data[1]['y'], good_classes)]
y = data[1]['y'][np.in1d(data[1]['y'], good_classes)]
y_dict = dict(zip(np.unique(y), range(len(np.unique(y)))))
y = np.vectorize(y_dict.get)(y)
split = train_test_split if n_train_per_class <= 0 else train_test_split_balanced
if n_train_per_class <= 0:
x_train, x_test, y_train, y_test = split(x,
y,
train_size=train_size,
test_size=test_size,
stratify=y)
x_train, x_val, y_train, y_val = split(x_train,
y_train,
train_size=0.8,
test_size=0.2,
stratify=y_train)
else:
n_train_per_class = int(np.round(n_train_per_class * 1.6))
x_train, x_test, y_train, y_test = split(
np.array(x),
np.array(y),
train_size=train_size,
test_size=test_size,
n_train_per_class=n_train_per_class,
n_test_per_class=n_test_per_class)
x_train, x_val, y_train, y_val = split(
x_train,
y_train,
train_size=0.8,
n_train_per_class=n_train_per_class,
test_size=0.2)
return (x_train, y_train), (x_val, y_val), (x_test, y_test)
import handshape_datasets
import sklearn
import keras
from keras.models import Model
dataset_id = "lsa16" #example for the lsa16 dataset
ver = "color" #version is optional argument, some datasets has one version
supr = False #supr is optional, some datasets can delete temporary files if it have .npz file
epochs = 15
batch_size = 64
dataset = handshape_datasets.load(dataset_id, version=ver,
delete=supr) #load the dataset
input_shape = dataset[0][0].shape #obtain the shape
classes = dataset[1]['y'].max() + 1 #obtain the number ofclasses
"""build the model"""
base_model = keras.applications.mobilenet.MobileNet(
input_shape=(input_shape[0], input_shape[1], 3),
weights='imagenet',
include_top=False)
output = keras.layers.GlobalAveragePooling2D()(base_model.output)
output = keras.layers.Dense(32, activation='relu')(output)
output = keras.layers.Dense(classes, activation='softmax')(output)
model = Model(inputs=base_model.input, outputs=output)
model.compile(optimizer='Adam',
loss='sparse_categorical_crossentropy',
metrics=['accuracy'])
"""split the dataset (its optional)"""
test_size = 0.1
X_train, X_test, Y_train, Y_test = sklearn.model_selection.train_test_split(
import handshape_datasets as hd
import logging
import numpy as np
#hd.list_datasets()
x, metadata = hd.load("rwth")
print(x.shape)
for k in metadata:
print(k, metadata[k].shape, metadata[k].min(), metadata[k].max())
#hd.clear("Nus1")
logging.basicConfig(format='%(levelname)s:%(message)s', level=logging.DEBUG)
#logging.debug(f"This message should go to the log file")
#logging.info("So should this")
#logging.warning("And this, too")
# Basic example showing how to get a dataset
import handshape_datasets as hd
DATASET_NAME = "PugeaultASL_A"
#ersion=dict({'1':'WithGabor'})
#ciarp_info = hd.info(DATASET_NAME)
#x,metadata = hd.load(DATASET_NAME,version='WithGabor') ciarp
#x,metadata = hd.load(DATASET_NAME, version='bw') nus1
#x,metadata = hd.load(DATASET_NAME, version='hn')
x, metadata = hd.load(DATASET_NAME)
print(x.shape)
print(x[1])
print(x.max())
print(x.min())
for k in metadata:
print(k, metadata[k].shape, metadata[k].min(), metadata[k].max())
#print(ciarp_info.summary())
#ciarp[0].show_dataset() #nunca devuelve un dataset, sino que en el caso de lsa16 devuelve un np.array y un dict (x y metadata)
# ciarp.show_dataset(subsets=["test_Kinect_WithGabor"],samples=128)
# ciarp.show_dataset(subsets=["test_Kinect_WithGabor"],samples=[1,2,3,0,15,1,200])
def load_ciarp(data_dir, config, splits):
"""
Load ciarp dataset.
Args:
data_dir (str): path of the directory with 'splits', 'data' subdirs.
config (dict): general dict with program settings.
splits (list): list of strings 'train'|'val'|'test'
Returns (dict): dictionary with keys as splits and values as tf.Dataset
"""
DATASET_NAME = "ciarp"
DATASET_PATH = "/develop/data/ciarp/data"
data = hd.load(DATASET_NAME, DATASET_PATH)
x_train, y_train = data['train_Kinect_WithoutGabor']
x_test, y_test = data['test_Kinect_WithoutGabor']
X = np.concatenate((x_train, x_test), axis=0)
y = np.concatenate((y_train, y_test), axis=0)
_, uniqueClasses = np.unique(y, return_counts=True)
_, amountPerTrain = np.unique(y_train, return_counts=True)
_, amountPerTest = np.unique(y_test, return_counts=True)
x_train, x_test, y_train, y_test = train_test_split_balanced(X,
y,
train_size=config['data.train_size'],
test_size=config['data.test_size'])
x_train, x_test = x_train / 255.0, x_test / 255.0
_, amountPerTrain = np.unique(y_train, return_counts=True)
_, amountPerTest = np.unique(y_test, return_counts=True)
train_datagen_args = dict(featurewise_center=True,
featurewise_std_normalization=True,
rotation_range=config['data.rotation_range'],
width_shift_range=config['data.width_shift_range'],
height_shift_range=config['data.height_shift_range'],
horizontal_flip=config['data.horizontal_flip'],
fill_mode='constant',
cval=0)
train_datagen = ImageDataGenerator(train_datagen_args)
train_datagen.fit(x_train)
test_datagen_args = dict(featurewise_center=True,
featurewise_std_normalization=True,
fill_mode='constant',
cval=0)
test_datagen = ImageDataGenerator(test_datagen_args)
test_datagen.fit(x_train)
w, h, c = list(map(int, config['model.x_dim'].split(',')))
ret = {}
for split in splits:
# n_way (number of classes per episode)
if split in ['val', 'test']:
n_way = config['data.test_way']
else:
n_way = config['data.train_way']
# n_support (number of support examples per class)
if split in ['val', 'test']:
n_support = config['data.test_support']
else:
n_support = config['data.train_support']
# n_query (number of query examples per class)
if split in ['val', 'test']:
n_query = config['data.test_query']
else:
n_query = config['data.train_query']
if split in ['val', 'test']:
y = y_test
x = x_test
dg = train_datagen
dg_args = train_datagen_args
else:
y = y_train
x = x_train
dg = test_datagen
dg_args = test_datagen_args
amountPerClass = amountPerTest if split in ['val', 'test'] else amountPerTrain
i = np.argsort(y)
x = x[i, :, :, :]
if config['model.type'] in ['processed']:
for index in i:
x[index, :, :, :] = dg.apply_transform(x[index], dg_args)
data = np.reshape(x, (len(uniqueClasses), amountPerClass[0], w, h, c))
data_loader = DataLoader(data,
n_classes=len(uniqueClasses),
n_way=n_way,
n_support=n_support,
n_query=n_query,
x_dim=(w, h, c))
ret[split] = data_loader
return ret