def main():
# add argumentation
parser = argparse.ArgumentParser(description='MobileNet_v2_DeepLab_v3 Pytorch Implementation')
#todo maybe make it work with multiple datasets?
#parser.add_argument('--dataset', default='cityscapes', choices=['cityscapes', 'other'],
# help='Dataset used in training MobileNet v2+DeepLab v3')
parser.add_argument('--root', default='./data/cityscapes', help='Path to your dataset')
parser.add_argument('--epoch', default=None, help='Total number of training epoch')
parser.add_argument('--lr', default=None, help='Base learning rate')
parser.add_argument('--pretrain', default=None, help='Path to a pre-trained backbone model')
parser.add_argument('--resume_from', default=None, help='Path to a checkpoint to resume model')
parser.add_argument('--logdir', default=None, help='Directory to save logs for Tensorboard')
parser.add_argument('--batch_size', default=128, help='Batch size for training')
args = parser.parse_args()
params = Params()
# parse args
if not os.path.exists(args.root):
if params.dataset_root is None:
raise ValueError('ERROR: Root %s doesn\'t exist!' % args.root)
else:
params.dataset_root = args.root
if args.epoch is not None:
params.num_epoch = int(args.epoch)
if args.lr is not None:
params.base_lr = args.lr
if args.pretrain is not None:
params.pre_trained_from = args.pretrain
if args.resume_from is not None:
params.resume_from = args.resume_from
if args.logdir is not None:
params.logdir = args.logdir
params.summary_dir, params.ckpt_dir = create_train_dir(params.logdir)
params.train_batch = int(args.batch_size)
LOG('Network parameters:')
print_config(params)
# create dataset and transformation
LOG('Creating Dataset and Transformation......')
datasets = create_dataset(params)
LOG('Creation Succeed.\n')
# create model
LOG('Initializing MobileNet and DeepLab......')
net = MobileNetv2_DeepLabv3(params, datasets)
LOG('Model Built.\n')
# let's start to train!
net.Train()
net.Test()
def __init__(self):
self.p = Printer(1)
self.param = Params()
self.m = Math()
self.am = AudioManager()
self.paths = Paths()
self.trainingDesc, self.testingDesc = self.scanForAudioFiles()
def gqn_draw_identity_model_fn(features, labels, mode, params=None):
_CONFIG['CONTEXT_SIZE'] = 1
params = Params(**_CONFIG)
_SEQ_LENGTH = params.SEQ_LENGTH
query_pose = features.query_camera
target_frame = labels
poses = features.query_camera
frames = labels
mu_target, ep_gqn = gqn_draw(query_pose, target_frame, poses, frames,
params, mode != tf.estimator.ModeKeys.PREDICT)
sigma_target = _linear_noise_annealing(params)
target_normal = tf.distributions.Normal(loc=mu_target, scale=sigma_target)
target_sample = tf.identity(target_normal.sample, name='target_sample')
l2_reconstruction = tf.identity(
tf.metrics.mean_squared_error(labels=target_frame,
predictions=mu_target))
if params.DEBUG:
tf.summary.image('context_frame_1', frames, max_outputs=1)
tf.summary.image('target_images', labels, max_outputs=1)
tf.summary.image('target_means', mu_target, max_outputs=1)
tf.summary.scalar('l2_reconstruction', l2_reconstruction[1])
gs = debug_canvas_image_mean(
[ep_gqn['canvas_{}'.format(i)] for i in range(_SEQ_LENGTH)])
tf.summary.image('generator_sequence_mean', gs, max_outputs=1)
if mode == tf.estimator.ModeKeys.PREDICT:
predictions = {'target_sample': target_sample}
estimator_spec = tf.estimator.EstimatorSpec(mode=mode,
predictions=predictions)
else:
mu_q, sigma_q, mu_pi, sigma_pi = [], [], [], []
for i in range(_SEQ_LENGTH):
mu_q.append(ep_gqn['mu_q_{}'.format(i)])
sigma_q.append(ep_gqn['sigma_q_{}'.format(i)])
mu_pi.append(ep_gqn['mu_pi_{}'.format(i)])
sigma_pi.append(ep_gqn['sigma_pi_{}'.format(i)])
elbo, ep_elbo = gqn_draw_elbo(mu_target, sigma_target, mu_q, sigma_q,
mu_pi, sigma_pi, target_frame)
if params.DEBUG:
tf.summary.scalar('target_llh', ep_elbo['target_llh'])
tf.summary.scalar('kl_regularizer', ep_elbo['kl_regularizer'])
if mode == tf.estimator.ModeKeys.TRAIN:
lr = _linear_lr_annealing(params)
optimizer = tf.train.AdamOptimizer(learning_rate=lr)
train_op = optimizer.minimize(elbo, tf.train.get_global_step())
estimator_spec = tf.estimator.EstimatorSpec(mode=mode,
loss=elbo,
train_op=train_op)
else:
eval_metric_ops = {
'l2_reconstruction':
tf.metrics.mean_squared_error(target_frame, mu_target)
}
estimator_spec = tf.estimator.EstimatorSpec(
mode=mode, loss=elbo, eval_metric_ops=eval_metric_ops)
return estimator_spec
def __init__(self):
self.paths = Paths();
self.param = Params();
self.pc = PrintConfig();
self.p = Printer(1);
self.am = AudioManager();
self.m = Math();
self.pickle = Pickle(self.paths.pickle, lTag=self.paths.tag1, sTag=self.paths.tag2);
self.data = self.pickle.LoadData();
def main():
# add argumentation
parser = argparse.ArgumentParser(
description='MobileNet_v2_DeepLab_v3 Pytorch Implementation')
parser.add_argument(
'--dataset',
default='cityscapes',
choices=['cityscapes', 'other'],
help='Dataset used in training MobileNet v2+DeepLab v3')
parser.add_argument('--root',
default='./data/cityscapes',
help='Path to your dataset')
parser.add_argument('--epoch',
default=None,
help='Total number of training epoch')
parser.add_argument('--lr', default=None, help='Base learning rate')
parser.add_argument('--pretrain',
default=None,
help='Path to a pre-trained backbone model')
parser.add_argument('--resume_from',
default=None,
help='Path to a checkpoint to resume model')
args = parser.parse_args()
params = Params()
# parse args
if not os.path.exists(args.root):
if params.dataset_root is None:
raise ValueError('ERROR: Root %s not exists!' % args.root)
else:
params.dataset_root = args.root
if args.epoch is not None:
params.num_epoch = args.epoch
if args.lr is not None:
params.base_lr = args.lr
if args.pretrain is not None:
params.pre_trained_from = args.pretrain
if args.resume_from is not None:
params.resume_from = args.resume_from
LOG('Network parameters:')
print_config(params)
# create dataset and transformation
LOG('Creating Dataset and Transformation......')
datasets = create_dataset(params)
LOG('Creation Succeed.\n')
# create model
LOG('Initializing MobileNet and DeepLab......')
net = MobileNetv2_DeepLabv3(params, datasets)
LOG('Model Built.\n')
# let's start to train!
net.Train()
net.Test()
def __init__(self):
self.paths = Paths()
self.param = Params()
self.pc = PrintConfig()
self.data = SpeachData()
self.p = Printer(1)
self.am = AudioManager()
self.m = Math()
self.pickle = Pickle(self.paths.pickle, sTag=self.paths.tag1)
self.data.raw = self.am.readAudio(self.paths.file)
def create_expansionai_env(env_id, video=False, params=Params()):
gym.spec(env_id)._kwargs = {
'armies': params.armies,
'board_size': params.board_size,
'offset_x': 0,
'offset_y': 0
}
env = gym.make(env_id)
if video:
env = wrappers.Monitor(env, 'test', force=True, mode='training')
return env
def main():
parser = argparse.ArgumentParser(
description='MobileNet_V2 Pytorch Implementation')
parser.add_argument('--dataset',
default='cifar10',
choices=['imagenet', 'cifar10', 'cifar100', 'other'],
help='Dataset used in training MobileNet V2')
parser.add_argument('--root',
default='./data/cifar10',
help='Path to your dataset')
args = parser.parse_args()
# parse args
if args.dataset == 'cifar10':
params = CIFAR10_params()
elif args.dataset == 'cifar100':
params = CIFAR100_params()
else:
params = Params()
params.dataset_root = args.root
if not os.path.exists(args.root):
print('ERROR: Root %s not exists!' % args.root)
exit(1)
""" TEST CODE """
# params = CIFAR100_params
# params.dataset_root = '/home/ubuntu/cifar100'
# create model
print('\nInitializing MobileNet......')
net = MobileNetv2(params)
print('Initialization Done.\n')
# create dataset and transformation
print('Loading Data......')
dataset = create_dataset(params)
print('Data Loaded.\n')
# let's start to train!
net.train_n_epoch(dataset)
def __init__(self):
self.paths = Paths()
self.param = Params()
self.pc = PrintConfig()
self.am = AudioManager()
self.p = Printer(1)
self.S = Synthesizer()
self.pickle = Pickle(self.paths.pickle)
self.decoded, self.original, self.coded = self.loadAll()
self.cP, self.cG, self.cLpc = self.organize()
self.cSn = self.SynthAll()
def __init__(self):
self.paths = Paths()
self.param = Params()
self.pc = PrintConfig()
self.p = Printer(1)
self.am = AudioManager()
self.m = Math()
self.pickle = Pickle(self.paths.pickle,
lTag=self.paths.tag4,
sTag=self.paths.tag5)
self.cc = CodeConfig()
self.cu = CodingUtils()
self.encoded = self.pickle.LoadEncoded()
def run():
params = Params()
grid_data = get_data(params.data_params)
data_dict = dataset_split(grid_data,
params.data_params['test_ratio'],
params.data_params['val_ratio'])
batch_gens = create_generator(data_dict, params.model_params['TRAJGRU']['batch_params'])
# for batch_idx, (x, y) in enumerate(batch_gens['train'].batch_next()):
# print(batch_idx, x.shape)
trained_model = trainer(batch_gens, **params.model_params['TRAJGRU'])
print('Training finished, saving the model')
model_file = open('results/conv_lstm.pkl', 'wb')
pickle.dump(trained_model, model_file)
def predict():
params = Params()
model_path = 'checkpoints/Checkpoint_epoch_150.pth.tar'
model = MobileNetv2_DeepLabv3Model(params).to(device)
checkpoint = torch.load(model_path)
state_dict = checkpoint['state_dict']
new_dict = {}
for k in state_dict.keys():
# print('K: {}, '.format(k))
new_dict['model.' + k] = state_dict[k]
model.load_state_dict(new_dict)
model.eval()
print('Model loaded.')
img_fs = [
'images/berlin_000004_000019_leftImg8bit.png',
'images/berlin_000002_000019_leftImg8bit.png',
]
transform = transforms.Compose(
[transforms.Resize(image_size),
transforms.ToTensor()])
for img_f in img_fs:
img = cv2.imread(img_f)
inp = Variable(transform(Image.fromarray(img)).to(device).unsqueeze(0))
print(inp.size())
out = model(inp)
print(out.size())
_, predictions = torch.max(out.data, 1)
prediction = predictions.cpu().numpy()[0]
print(prediction)
mask_color = np.asarray(label_to_color_image(prediction, 'cityscapes'),
dtype=np.uint8)
frame = cv2.resize(img, (1024, 512))
print('msk: {}, frame: {}'.format(mask_color.shape, frame.shape))
res = cv2.addWeighted(frame, 0.5, mask_color, 0.7, 1)
cv2.imshow('res', res)
while True:
cv2.waitKey(27)
def __init__(self, num_inputs, action_space, params=Params()):
""" Initis Actor Critic """
super(ActorCritic, self).__init__()
self.params = params
logger.info("Init ActorCritic num_inputs: %s and action_space: %s" %
(num_inputs, action_space))
hidden_layer_neuron_size = self.params.board_size * self.params.board_size
self.fc1 = nn.Linear(self.params.board_size, hidden_layer_neuron_size)
self.fc2 = nn.Linear(hidden_layer_neuron_size,
hidden_layer_neuron_size)
self.fc3 = nn.Linear(hidden_layer_neuron_size, num_inputs)
self.lstm = nn.LSTMCell(
self.params.board_dimension_size * self.params.board_size,
self.params.lstm_size)
# num_outputs = int(action_space.high[0]) # action_space.num_discrete_space
num_outputs = int(action_space.n) # action_space.num_discrete_space
self.critic_linear = nn.Linear(256, 1) # output = V(S)
self.actor_linear = nn.Linear(256, num_outputs) # output = Q(S, a)
self.apply(init_weights)
self.actor_linear.weight.data = normalized_columns_initializer(
self.actor_linear.weight.data, 0.01)
self.actor_linear.bias.data.fill_(0)
self.critic_linear.weight.data = normalized_columns_initializer(
self.critic_linear.weight.data, 0.01)
self.critic_linear.bias.data.fill_(0)
self.lstm.bias_ih.data.fill_(0)
self.lstm.bias_hh.data.fill_(0)
self.train()
#基础程序设置
seed=789
torch.manual_seed(seed)#seed: long
np.random.seed(seed)
torch.cuda.manual_seed(seed)
#设备信息
print('GPU available: ',torch.cuda.is_available())
print('CUDNN available: ',torch.backends.cudnn.enabled)
print('GPU number: ',torch.cuda.device_count())
# parser = OptionParser()
# parser.add_option("--train", dest="trainFile",
# default="", help="train dataset")
# parser.add_option("--dev", dest="devFile",
# default="", help="dev dataset")
# parser.add_option("--test", dest="testFile",
# default="", help="test dataset")
# (options, args) = parser.parse_args()
params=Params()
if not torch.cuda.is_available() or not torch.backends.cudnn.enabled or torch.cuda.device_count()<=0:
params.use_gpu=False
params.device=torch.device("cuda:"+str(params.device_id) if params.use_gpu else 'cpu')
params.show()
master=Nebuchadnezzar(params)
master.data_process()
master.train_process()
# Create the Timeline object, and write it to a json file
fetched_timeline = timeline.Timeline(run_metadata.step_stats)
chrome_trace = fetched_timeline.generate_chrome_trace_format()
with open(os.path.join(params.log_dir, 'timeline_01.json'), 'w') as f:
f.write(chrome_trace)
with open(os.path.join(params.log_dir, 'mem_info.json'), 'w') as f:
f.write(str(run_metadata))
else:
loss_value, _ = sess.run([loss, train_op], feed_dict=feed_dict)
if step % params.ckpt_save_steps == 0:
model.save(saver, sess, params.log_dir, step)
duration = time.time() - start_time
print('step {:d} \t loss = {:.3f}, ({:.3f} sec/step)'.format(step, loss_value, duration))
coord.request_stop()
coord.join(threads)
if __name__ == '__main__':
args = get_arguments()
params = Params()
params = load_json_to_params(params, args.json_path)
params.dataset_directory = '/home/ddegeus/datasets/Cityscapes/training/'
params.filelist_filepath = '/home/ddegeus/datasets/Cityscapes/training/panoptic/filenames.lst'
params.is_training = True
params.batch_norm_istraining = True
print(params)
train(params)
out_file = np.stack([label_ids, instance_ids], axis=2)
print(np.min(label_ids), np.max(label_ids))
save_dir = params.save_dir
im_name_base = os.path.splitext(os.path.basename(str(filename)))[0]
out_fname = os.path.join(save_dir, im_name_base + '.png')
print(out_fname)
print(np.min(out_file[..., 0]), np.max(out_file[..., 0]))
Image.fromarray(out_file.astype(np.uint8)).save(out_fname)
if __name__ == '__main__':
args = get_arguments()
params = Params()
params = load_json_to_params(params, args.json_path)
params.num_steps_predict = params.num_steps_eval
params.save_predictions = args.save_predictions
params.save_dir = args.save_dir
params.is_training = False
params.batch_norm_istraining = False
params.height_input = 512
params.width_input = 1024
params.height_orig = 604
params.width_orig = 960
params.Nb = 1
filenames_list = list()
for file in os.listdir(args.image_dir):
time_str = datetime.datetime.now().isoformat()
if verbose:
if random.randint(0, 10) < 1:
print(("%s: DIS step %d, loss %f with acc %f " %
(time_str, step, current_loss, accuracy)))
def getReward(self, batch, sess, verbose=True):
feed_dict = {
self.input_x_1: batch[:, 0],
self.input_x_2: batch[:, 1],
self.input_x_3: batch[:, 2],
self.dropout_keep_prob_holder: 1.0
}
scores = sess.run(self.reward, feed_dict)
return scores
if __name__ == "__main__":
from config import Params
opts = Params()
opts.parseArgs()
from dataHelper import InsuranceQA
dataset = InsuranceQA(opts)
model = Discriminator(opts)
sess = tf.Session()
sess.run(tf.global_variables_initializer())
batch = next(iter(dataset.generate_uniform_pair()))
from envs import create_expansionai_env
from model import ActorCritic
from test import test
from train import train
LOGGING_FORMAT = '%(asctime)s - %(name)s - %(thread)d|%(process)d - %(levelname)s - %(message)s'
logging.basicConfig(format=LOGGING_FORMAT)
# logging.getLogger('Model').setLevel(logging.INFO)
# logging.getLogger('ExpansionAiEnv').setLevel(logging.DEBUG)
logging.getLogger('Train').setLevel(logging.INFO)
logging.getLogger('Test').setLevel(logging.INFO)
# Main run
os.environ['OMP_NUM_THREADS'] = '1' # 1 thread per core
params = Params(
) # creating the params object from the Params class, that sets all the model parameters
params.max_episode_length = 1_000_000
params.num_processes = 3
torch.manual_seed(params.seed) # setting the seed (not essential)
env = create_expansionai_env(
params.env_name,
params) # we create an optimized environment thanks to universe
# shared_model is the model shared by the different agents (different threads in different cores)
shared_model = ActorCritic(env.observation_space.shape[0], env.action_space)
# storing the model in the shared memory of the computer, which allows the threads to have access to this shared memory even if they are in different cores
shared_model.share_memory()
# the optimizer is also shared because it acts on the shared model
optimizer = my_optim.SharedAdam(shared_model.parameters(), lr=params.lr)
transforms.Compose([
RandomResizedCrop(params.image_size, scale=(0.5, 2.0)),
RandomHorizontalFlip(),
ToTensor()
]),
'val':
transforms.Compose([
RandomResizedCrop(params.image_size, scale=(0.5, 2.0)),
ToTensor()
]),
'test':
transforms.Compose([ToTensor()])
}
# file_dir = {p: os.path.join(params.dataset_root, p) for p in phase}
# datasets = {Cityscapes(file_dir[p], mode=p, transforms=transform[p]) for p in phase}
datasets = {
p: Cityscapes(params.dataset_root, mode=p, transforms=transform[p])
for p in phase
}
return datasets
if __name__ == '__main__':
from config import Params
pp = Params()
pp.dataset_root = '/media/ubuntu/disk/cityscapes'
datasets = create_datasets(pp)
from optparse import OptionParser
from config import Params
from Pythagoras_master import Pythagoras
if __name__ == "__main__":
parser = OptionParser()
parser.add_option("--train",
dest="trainFile",
default="",
help="train dataset")
parser.add_option("--dev", dest="devFile", default="", help="dev dataset")
parser.add_option("--test",
dest="testFile",
default="",
help="test dataset")
(options, args) = parser.parse_args()
params = Params()
if options.trainFile != "":
params.process_params.trainFile = options.trainFile
if options.devFile != "":
params.process_params.devFile = options.devFile
if options.testFile != "":
params.process_params.testFile = options.testFile
master = Pythagoras(params)
master.train()
edges_invert = edges_invert.astype(np.uint8)
class_colors = class_colors.astype(np.uint8) * np.expand_dims(edges_invert, axis=2) + np.expand_dims(
edges_total, axis=2)
img_obj = Image.fromarray(np.uint8(class_colors))
ax.imshow(img_obj)
plt.waitforbuttonpress(timeout=5)
if __name__ == '__main__':
args = get_arguments()
params = Params()
params = load_json_to_params(params, args.json_path)
params.dataset_directory = '/home/ddegeus/datasets/Cityscapes/validation/'
params.filelist_filepath = '/home/ddegeus/datasets/Cityscapes/validation/panoptic/filenames.lst'
params.is_training = False
params.batch_norm_istraining = False
params.num_steps_predict = params.num_steps_eval
params.height_input = 512
params.width_input = 1024
params.Nb = 1
predict(params)