def main(argv):
if not os.path.exists("./experiments/" + FLAGS.exp_name):
os.makedirs("./experiments/" + FLAGS.exp_name)
# Image mode
if FLAGS.img_mode == 'rgb':
img_channels = 3
elif FLAGS.img_mode == 'grayscale':
img_channels = 1
else:
raise IOError("Unidentified image mode: use 'grayscale' or 'rgb'")
# Output dimension (one for steering and one for collision)
output_dim = 5
train_datagen = utils.NavemDataGenerator(rescale=1. / 255)
train_generator = train_datagen.flow_from_directory(
os.path.sep.join(["./datasets", FLAGS.dataset, "train"]),
shuffle=True,
color_mode=FLAGS.img_mode,
target_size=(FLAGS.img_width, FLAGS.img_height),
crop_size=(FLAGS.crop_img_width, FLAGS.crop_img_height),
batch_size=FLAGS.batch_size)
val_datagen = utils.NavemDataGenerator(rescale=1. / 255)
val_generator = val_datagen.flow_from_directory(
os.path.sep.join(["./datasets", FLAGS.dataset, "val"]),
shuffle=True,
color_mode=FLAGS.img_mode,
target_size=(FLAGS.img_width, FLAGS.img_height),
crop_size=(FLAGS.crop_img_width, FLAGS.crop_img_height),
batch_size=FLAGS.batch_size)
# Weights to restore
weights_path = os.path.join("./experiments/" + FLAGS.exp_name,
FLAGS.weights_fname)
initial_epoch = 0
if not FLAGS.restore_model:
# In this case weights will start from random
weights_path = None
else:
# In this case weigths will start from the specified model
initial_epoch = FLAGS.initial_epoch
# Define model
model = models.getModel(FLAGS.img_width, FLAGS.img_height, img_channels,
output_dim, weights_path)
# Serialize model into json
json_model_path = os.path.join("./experiments/" + FLAGS.exp_name,
FLAGS.json_model_fname)
utils.modelToJson(model, json_model_path)
# Train model
trainModel(train_generator, val_generator, model, initial_epoch)
def main(argv):
#python3 evaluation.py --exp_name="exp_005" --dataset="dataset_navem_224_224" --weights_fname="weights_050.h5" --batch_size=64
#python3 evaluation.py --exp_name="exp_029" --dataset="\vgg16\sidewalk_accx" --weights_fname="model_weights_259.h5" --batch_size=64
# Set testing mode (dropout/batchnormalization)
########### command aws #############
#python3 evaluation.py --exp_name="exp_029" --dataset="sidewalk_accy_proportion_classes_20" --weights_fname="model_weights_259.h5" --batch_size=64
TEST_PHASE = 0
TRAIN_PHASE = 1
K.set_learning_phase(TEST_PHASE)
# Generate testing data
test_datagen = utils.NavemDataGenerator(rescale=1. / 255)
test_generator = test_datagen.flow_from_directory(
os.path.sep.join([
".\\..\\datasets", "vgg16", FLAGS.dataset, FLAGS.dataset, "train"
]),
shuffle=False,
color_mode=FLAGS.img_mode,
target_size=(FLAGS.img_width, FLAGS.img_height),
crop_size=(FLAGS.crop_img_height, FLAGS.crop_img_width),
batch_size=FLAGS.batch_size)
# Load json and create model
json_model_path = os.path.join(".\\..\\experiments", FLAGS.exp_name,
FLAGS.json_model_fname)
model = utils.jsonToModel(json_model_path)
# Load weights
weights_load_path = os.path.join("./experiments/" + FLAGS.exp_name,
FLAGS.weights_fname)
try:
model.load_weights(weights_load_path)
print("Loaded model from {}".format(weights_load_path))
except:
print("Impossible to find weight path. Returning untrained model")
# Compile model
model.compile(loss='mse',
optimizer='adam') # change to "categorical_crossentropy"
# Get predictions and ground truth
n_samples = test_generator.samples
nb_batches = int(np.ceil(n_samples / FLAGS.batch_size))
predictions, ground_truth, t = utils.compute_predictions_and_gt(
model, test_generator, nb_batches, verbose=1)
print(predictions[1])
print(ground_truth[1])
def main(argv):
if not os.path.exists("./experiments/" + FLAGS.exp_name):
os.makedirs("./experiments/" + FLAGS.exp_name)
# Image mode
if FLAGS.img_mode=='rgb':
img_channels = 3
elif FLAGS.img_mode == 'grayscale':
img_channels = 1
else:
raise IOError("Unidentified image mode: use 'grayscale' or 'rgb'")
# Output dimension (one for steering and one for collision)
output_dim = 1
print("[INFO] loading attributes...")
##inputPath = os.path.sep.join(["./datasets/labels", FLAGS.dataset + ".txt"])
##df = datasets.load_navem_attributes(inputPath)
# load the house images and then scale the pixel intensities to the range [0, 1]
print("[INFO] loading images...")
##images = datasets.load_navem_images(df)
##images = images / 255.0
# partition the data into training and testing splits using 75% of
# the data for training and the remaining 25% for testing
#print("\n\n\n" + str(len(df)) + "| " + str(len(images)) + "\n\n\n")
##split = train_test_split(df, images, test_size=0.20, random_state=42)
#programPause = input("Press the <ENTER> key to continue save train and test...")
##(trainAttrX, testAttrX, trainImagesX, testImagesX) = split
#Save train and test state
##utils.write_divided_dataset(os.path.join("./experiments/" + FLAGS.exp_name), trainAttrX, testAttrX)
#Nomalization
#trainY = (trainAttrX["gyro_z"] - trainAttrX["gyro_z"].min()) / (trainAttrX["gyro_z"].max() - trainAttrX["gyro_z"].min())
#testY = (testAttrX["gyro_z"] - testAttrX["gyro_z"].min()) / (testAttrX["gyro_z"].max() - testAttrX["gyro_z"].min())
##train_generator = (trainImagesX, trainAttrX["gyro_z"])
##val_generator = (testImagesX, testAttrX["gyro_z"])
#Data augmentation FLAGS
#rotation_range = 0.2
#width_shift_range = 0.2
#height_shift_range=0.2
train_datagen = utils.NavemDataGenerator(rescale = 1./255, rotation_range = 0.2, width_shift_range = 0.2, height_shift_range=0.2)
train_generator = train_datagen.flow_from_directory(os.path.sep.join(["./datasets", FLAGS.dataset, "train"]),
shuffle = True,
color_mode=FLAGS.img_mode,
target_size=(FLAGS.img_width, FLAGS.img_height),
crop_size=(FLAGS.crop_img_width, FLAGS.crop_img_height),
batch_size = FLAGS.batch_size)
#print(train_generator.samples)
val_datagen = utils.NavemDataGenerator(rescale=1./255, rotation_range = 0.2, width_shift_range = 0.2, height_shift_range=0.2)
#val_datagen.fit(testImagesX)
val_generator = val_datagen.flow_from_directory(os.path.sep.join(["./datasets", FLAGS.dataset, "val"]),
shuffle = True,
color_mode=FLAGS.img_mode,
target_size=(FLAGS.img_width, FLAGS.img_height),
crop_size=(FLAGS.crop_img_width, FLAGS.crop_img_height),
batch_size = FLAGS.batch_size)
# Weights to restore
weights_path = os.path.join("./experiments/" + FLAGS.exp_name, FLAGS.weights_fname)
initial_epoch = 0
if not FLAGS.restore_model:
# In this case weights will start from random
weights_path = None
else:
# In this case weigths will start from the specified model
initial_epoch = FLAGS.initial_epoch
# Define model
model = models.getModel(FLAGS.img_width, FLAGS.img_height, img_channels, output_dim, weights_path)
#model = models.create_cnn(64, 64, 3, regress=True)
# Serialize model into json
json_model_path = os.path.join("./experiments/" + FLAGS.exp_name, FLAGS.json_model_fname)
utils.modelToJson(model, json_model_path)
# Train model
trainModel(train_generator, val_generator, model, initial_epoch)