def train(config):
#####################
## Dataset
is_2d_model = config['model'] in ("cnn2d", "vgg16", "vgg19", "inception")
dataset_train = GestureDataset(config["dataset_path"],
which_split='train',
last_frame_only=is_2d_model,
batch_size=config["batch_size"])
dataset_validation = GestureDataset(config["dataset_path"],
which_split='validation',
last_frame_only=is_2d_model,
batch_size=config["batch_size"])
#####################
## Model
kwargs = dict(out_shape=config["out_shape"],
in_shape=config["in_shape"],
config=config)
if config['model'] == "cnn2d":
model = create_model_2d(**kwargs)
elif config['model'] == "cnn3d":
model = create_model_3d(**kwargs)
elif config['model'] == "vgg16":
model = create_model_vgg(**kwargs)
elif config['model'] == "vgg19":
model = create_model_vgg19(**kwargs)
elif config['model'] == "inception":
model = create_model_inception(**kwargs)
else:
raise ValueError("Unknown model {}".format(config["model"]))
model.summary()
#####################
## Data augmentation
dataset_train_augmented = default_training_preprocessing(
config, dataset_train)
dataset_validation_prepared = default_evaluation_preprocessing(
config, dataset_validation)
#####################
## Training setup
metrics.BATCH_SIZE = config["batch_size"]
model.compile(optimizer='adam', loss='mse', metrics=[motion_metric])
#####################
# Callbacks
filepath = get_checkpoint_filepath(config, pattern='weights.hdf5')
checkpoint_saver = ModelCheckpoint(
filepath,
monitor='val_motion_metric',
save_best_only=True, # only overwrite if model is better
mode='max' # higher is better for this metric
)
#####################
## Go!
history = model.fit_generator(
generator=dataset_train_augmented,
validation_data=dataset_validation_prepared,
callbacks=[checkpoint_saver],
epochs=config["epochs"],
verbose=2 # 0 = silent, 1 = progress bar, 2 = one line per epoch.
)
return model, history
def trainfforvalidationandtest(config):
#####################
## Dataset
is_2d_model = config['model'] in ("cnn2d", "vgg16", "vgg19", "inception")
dataset_train = GestureDataset(get_dataset_dir(config),
which_split='train',
last_frame_only=is_2d_model,
batch_size=config['batch_size'])
dataset_validation = GestureDataset(get_dataset_dir(config),
which_split='validation',
last_frame_only=is_2d_model,
batch_size=config['batch_size'])
#####################
## Model
kwargs = dict(out_shape=config["out_shape"],
in_shape=config["in_shape"],
config=config)
if config['model'] == "cnn2d":
model = create_model_2d(**kwargs)
elif config['model'] == "cnn3d":
model = create_model_3d(**kwargs)
elif config['model'] == "vgg16":
model = create_model_vgg(**kwargs)
elif config['model'] == "vgg19":
model = create_model_vgg19(**kwargs)
elif config['model'] == "inception":
model = create_model_inception(**kwargs)
else:
raise ValueError("Unknown model {}".format(config["model"]))
model.summary()
#####################
## Data augmentation
dataset_train_augmented = default_training_preprocessing(
config, dataset_train)
dataset_validation_prepared = default_evaluation_preprocessing(
config, dataset_validation)
#####################
## Training setup
metrics.BATCH_SIZE = config["batch_size"]
adam = keras.optimizers.Adam(lr=0.0001,
beta_1=0.9,
beta_2=0.999,
amsgrad=False)
model.compile(optimizer=adam, loss='mse', metrics=[motion_metric])
#####################
# Callbacks
filepath = get_checkpoint_filepath(config, pattern='weights.hdf5')
checkpoint_saver = ModelCheckpoint(
filepath,
monitor='val_motion_metric',
save_best_only=True, # only overwrite if model is better
mode='max' # higher is better for this metric
)
#####################
## Go!
# print("shape of train dataste augm:", dataset_train_augmented.shape)
history = model.fit_generator(
generator=dataset_train_augmented,
validation_data=dataset_validation_prepared,
callbacks=[checkpoint_saver],
epochs=config["epochs"],
verbose=2 # 0 = silent, 1 = progress bar, 2 = one line per epoch.
)
print(model.metrics_names)
x_evaluate = []
y_evaluate = []
for i in range(len(dataset_validation_prepared)):
batch, teachers = dataset_validation_prepared[i]
x_evaluate.append(batch)
y_evaluate.append(teachers)
x_eval = np.asarray(x_evaluate)
if is_2d_model:
x_eval = x_eval.reshape(-1, *x_eval.shape[-3:])
else:
x_eval = x_eval.reshape(-1, *x_eval.shape[-4:])
y_eval = np.asarray(y_evaluate)
y_eval = y_eval.reshape(-1, *y_eval.shape[-2:])
loss, motion_score = model.evaluate(x_eval,
y_eval,
batch_size=config['batch_size'])
return model, history, motion_score
def train(config):
#######Configs for GPU
from keras.backend.tensorflow_backend import set_session
import tensorflow as tf
configGraphics = tf.ConfigProto()
configGraphics.gpu_options.allow_growth = True # dynamically grow the memory used on the GPU
configGraphics.log_device_placement = True # to log device placement (on which device the operation ran)
sess = tf.Session(config=configGraphics)
set_session(
sess) # set this TensorFlow session as the default session for Keras
#####################
## Dataset
is_2d_model = config['model'] in ("cnn2d", "vgg16", "vgg19", "inception")
dataset_train = GestureDataset(get_dataset_dir(config),
which_split='train',
last_frame_only=is_2d_model,
batch_size=config["batch_size"])
dataset_validation = GestureDataset(get_dataset_dir(config),
which_split='validation',
last_frame_only=is_2d_model,
batch_size=config["batch_size"])
#####################
## Model
kwargs = dict(out_shape=config["out_shape"],
in_shape=config["in_shape"],
config=config)
if config['model'] == "cnn2d":
model = create_model_2d(**kwargs)
elif config['model'] == "cnn3d":
model = create_model_3d(**kwargs)
elif config['model'] == "vgg16":
model = create_model_vgg(**kwargs)
elif config['model'] == "vgg19":
model = create_model_vgg19(**kwargs)
elif config['model'] == "inception":
model = create_model_inception(**kwargs)
else:
raise ValueError("Unknown model {}".format(config["model"]))
model.summary()
#####################
## Data augmentation
dataset_train_augmented = default_training_preprocessing(
config, dataset_train)
dataset_validation_prepared = default_evaluation_preprocessing(
config, dataset_validation)
#####################
## Training setup
metrics.BATCH_SIZE = config["batch_size"]
model.compile(optimizer='sgd', loss='mse', metrics=[motion_metric])
#####################
# Callbacks
filepath = get_checkpoint_filepath(config, pattern='weights.hdf5')
checkpoint_saver = ModelCheckpoint(
filepath,
monitor='val_motion_metric',
save_best_only=True, # only overwrite if model is better
mode='max' # higher is better for this metric
)
#####################
## Go!
# print("shape of train dataste augm:", dataset_train_augmented.shape)
history = model.fit_generator(
generator=dataset_train_augmented,
validation_data=dataset_validation_prepared,
callbacks=[checkpoint_saver],
epochs=config["epochs"],
verbose=2 # 0 = silent, 1 = progress bar, 2 = one line per epoch.
)
return model, history
def train(params):
#####################
#Dataset
dataset_train = GestureDataset(dataset_dir,
which_split='train',
last_frame_only=True,
batch_size=params['batch_size'])
dataset_validation = GestureDataset(dataset_dir,
which_split='validation',
last_frame_only=True,
batch_size=params['batch_size'])
#####################
## Model
model = create_model_vgg19(params)
model.summary()
#####################
## Data augmentation
dataset_train_augmented = default_training_preprocessing(
config_dict, dataset_train)
dataset_validation_prepared = default_evaluation_preprocessing(
config_dict, dataset_validation)
#####################
## Training setup
metrics.BATCH_SIZE = params['batch_size']
model.compile(optimizer=params['optimizer'],
loss='mse',
metrics=[motion_metric])
#####################
# Callbacks
# filepath = get_checkpoint_filepath(config, pattern='weights.hdf5')
# checkpoint_saver = ModelCheckpoint(filepath,
# monitor='val_motion_metric',
# save_best_only=True, # only overwrite if model is better
# mode='max' # higher is better for this metric
# )
#####################
## Go!
# print("shape of train dataste augm:", dataset_train_augmented.shape)
history = model.fit_generator(
generator=dataset_train_augmented,
validation_data=dataset_validation_prepared,
#callbacks=[checkpoint_saver],
epochs=params['epochs'],
verbose=2 # 0 = silent, 1 = progress bar, 2 = one line per epoch.
)
print(model.metrics_names)
x_evaluate = []
y_evaluate = []
for i in range(len(dataset_validation_prepared)):
batch, teachers = dataset_validation_prepared[i]
x_evaluate.append(batch)
y_evaluate.append(teachers)
print("meine samples für evaluate als liste")
print(len(x_evaluate))
x_eval = np.asarray(x_evaluate)
print(x_eval.shape)
x_eval = x_eval.reshape(-1, *x_eval.shape[-3:])
print("meine samples als np array")
print(x_eval.shape)
print("meine teachers für evaluate als liste")
print(len(y_evaluate))
y_eval = np.asarray(y_evaluate)
print(y_eval.shape)
y_eval = y_eval.reshape(-1, *y_eval.shape[-2:])
print("meine teachers als np array")
print(y_eval.shape)
loss, motion_score = model.evaluate(x_eval,
y_eval,
batch_size=params['batch_size'])
print("mein score")
print(motion_score)
return model, history, motion_score