def gen(hwm, host, port):
for tup in client_generator(hwm=hwm, host=host, port=port):
X, Y, _ = tup
Y = Y[:, -1]
if X.shape[1] == 1: # no temporal context
X = X[:, -1]
yield X, Y
def gen(hwm, host, port):
for tup in client_generator(hwm=hwm, host=host, port=port):
X, Y, _ = tup
Y = Y[:, -1]
if X.shape[1] == 1: # no temporal context
X = X[:, -1]
yield X, Y
def gen(hwm, host, port):
for tup in client_generator(hwm=hwm, host=host, port=port):
X, Y, _ = tup
Y = Y[:, -1]
if X.shape[1] == 1: # no temporal context
X = X[:, -1]
#print(X.shape)
# X = np.swapaxes(X, 1, 3)
# X = np.array([cv2.resize(x, (40, 40), interpolation = cv2.INTER_AREA).flatten() for x in X])
# #X = np.swapaxes(X, 3, 1)
# #print(X.shape)
# X = X-128 / 128
yield X, Y
def gen(hwm, host, port):
for tup in client_generator(hwm=hwm, host=host, port=port):
Image, Steer, Speed, Gas, Gear, Brake = tup
if Image.shape[1] == 1: # no temporal context
Image = Image[:, -1]
#Steer = Steer[:, -1]
#Speed = Speed[:,-1]
Gas = Gas[:, -1]
Gear = Gear[:, -1]
Brake = Brake[:, -1]
yield [Image, Speed], Steer
else:
yield [
Image, Speed
], Steer #Need to change according to the modelling parameter
sess = tf.InteractiveSession(config=tfconfig)
tf.global_variables_initializer().run()
# saver
saver = tf.train.Saver(max_to_keep=40)
if config.pretrained_model_path is not None:
saver.restore(sess, config.pretrained_model_path)
print("Loaded the pretrained model: %s" %
(config.pretrained_model_path))
# Summary writer
summary_writer = tf.summary.FileWriter(args.log_path,
graph=tf.get_default_graph())
# Train over the dataset
data_train = client_generator(hwm=20, host="localhost", port=args.port)
data_val = client_generator(hwm=20, host="localhost", port=args.val_port)
curr_loss = 0
for i in range(config.epochsize):
feats_batch, angle_batch, speed_batch = next(data_train)
# Preprocessing
feats, curvatures, angles = pre_processor.process(
sess, feats_batch, angle_batch, speed_batch)
# Training
if config.use_curvature:
feed_dict = {VA_model.features: feats, VA_model.y: curvatures}
else:
feed_dict = {VA_model.features: feats, VA_model.y: angles}
g_train, d_train, sampler, saver, loader, [G, E, T] = get_model(
sess=sess, name=args.name, batch_size=args.batch, gpu=args.gpu)
print("loading weights...")
G.load_weights("./outputs/results_autoencoder/G_weights.keras".format(
args.name))
E.load_weights("./outputs/results_autoencoder/E_weights.keras".format(
args.name))
checkpoint_dir = './outputs/results_' + args.name
T.load_weights(checkpoint_dir + "/T_weights.keras")
if not os.path.exists("./video_" + args.name):
os.makedirs("./video_" + args.name)
# get data
data = client_generator(hwm=20, host="localhost", port=5557)
X = next(data)[0] # [:, ::2]
sh = X.shape
X = X.reshape((-1, 3, 160, 320))
X = np.asarray(
[cv2.resize(x.transpose(1, 2, 0), (160, 80)) for x in X])
X = X / 127.5 - 1.
x = X.reshape((sh[0], args.time, 80, 160, 3))
# estimate frames
z_dim = 512
I = E.input
E_out = E(I)
O = G.input
G_out = G(O)
print "Sampling..."
def gen(hwm, host, port):
for tup in client_generator(hwm=hwm, host=host, port=port):
X = cleanup(tup)
yield X
def gen(hwm, host, port):
for tup in client_generator(hwm=hwm, host=host, port=port):
X = cleanup(tup)
yield X
os.makedirs(samples_path)
if not os.path.exists("./outputs/logs_" + args.name):
os.system("rm -rf ./outputs/logs_" + args.name)
#gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=0.333)
#with tf.Session(config=tf.ConfigProto(gpu_options=gpu_options)) as sess:
with tf.Session() as sess:
ftrain, ftest, loader, saver, extras = get_model(sess=sess,
name=args.name,
batch_size=args.batch,
time_len=args.time)
# start from checkpoint
if args.loadweights:
loader()
train_gen = client_generator(hwm=20, host=args.host, port=args.port)
train_model(args.name,
ftrain,
train_gen,
samples_per_epoch=args.epochsize,
ftest=None,
validation_data=None,
test_data=None,
nb_val_samples=100,
nb_epoch=args.epoch,
verbose=1,
saver=saver,
gif=args.gif)
with tf.Session() as sess:
K.set_session(sess)
g_train, d_train, sampler, saver, loader, [G, E, T] = get_model(sess=sess, name=args.name, batch_size=args.batch, gpu=args.gpu)
print("loading weights...")
G.load_weights("./outputs/results_autoencoder/G_weights.keras".format(args.name))
E.load_weights("./outputs/results_autoencoder/E_weights.keras".format(args.name))
checkpoint_dir = './outputs/results_' + args.name
T.load_weights(checkpoint_dir+"/T_weights.keras")
if not os.path.exists("./video_"+args.name):
os.makedirs("./video_"+args.name)
# get data
data = client_generator(hwm=20, host="localhost", port=5557)
X = next(data)[0] # [:, ::2]
sh = X.shape
X = X.reshape((-1, 3, 160, 320))
X = np.asarray([cv2.resize(x.transpose(1, 2, 0), (160, 80)) for x in X])
X = X/127.5 - 1.
x = X.reshape((sh[0], args.time, 80, 160, 3))
# estimate frames
z_dim = 512
I = E.input
E_out = E(I)
O = G.input
G_out = G(O)
print "Sampling..."
for i in tqdm(range(128)):
