def main(_):
if not os.path.exists(args.ckpt_dir):
os.makedirs(args.ckpt_dir)
if not os.path.exists(args.sample_dir):
os.makedirs(args.sample_dir)
if not os.path.exists(args.test_dir):
os.makedirs(args.test_dir)
lr = args.lr * np.ones([args.epoch])
lr[30:] = lr[0] / 10.0
if args.use_gpu:
# added to control the gpu memory
print("GPU\n")
gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=0.9)
with tf.Session(config=tf.ConfigProto(
gpu_options=gpu_options)) as sess:
model = denoiser(sess, sigma=args.sigma)
if args.phase == 'train':
denoiser_train(model, lr=lr)
elif args.phase == 'test':
denoiser_test(model)
else:
print('[!]Unknown phase')
exit(0)
else:
print("CPU\n")
with tf.Session() as sess:
model = denoiser(sess, sigma=args.sigma)
if args.phase == 'train':
denoiser_train(model, lr=lr)
elif args.phase == 'test':
denoiser_test(model)
else:
print('[!]Unknown phase')
exit(0)
def main(_):
#the following string is attached to checkpoint, log and image folder names
eps_str = str(int(np.log10(args.eps)))
name = "DnCNN_" + args.cost + str(args.type) + "_sigma" + str(
int(args.sigma))
ckpt_dir = args.ckpt_dir + "/" + name
sample_dir = args.sample_dir + "/" + name
test_dir = args.test_dir + "/" + name
log_dir = args.log_dir + "/" + name
if not os.path.exists(args.ckpt_dir):
os.makedirs(args.ckpt_dir)
lr = args.lr * np.ones([args.epoch])
lr[40:] = lr[0] / 10.0 #lr decay
if args.use_gpu:
# added to control the gpu memory
print("GPU\n")
gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=0.9)
with tf.Session(config=tf.ConfigProto(
gpu_options=gpu_options)) as sess:
model = denoiser(sess,
sigma=args.sigma,
eps=args.eps,
cost_str=args.cost,
ckpt_dir=ckpt_dir,
sample_dir=sample_dir,
log_dir=log_dir)
if args.phase == 'train':
denoiser_train(model, lr=lr)
elif args.phase == 'test':
denoiser_test(model, test_dir)
else:
print('[!]Unknown phase')
exit(0)
else:
print("CPU\n")
with tf.Session() as sess:
model = denoiser(sess,
sigma=args.sigma,
eps=args.eps,
cost_str=args.cost,
ckpt_dir=ckpt_dir,
sample_dir=sample_dir,
log_dir=log_dir)
if args.phase == 'train':
denoiser_train(model, lr=lr)
elif args.phase == 'test':
denoiser_test(model, test_dir)
else:
print('[!]Unknown phase')
exit(0)
def main(_):
# checkpoint dir
if not os.path.exists(args.ckpt_dir):
os.makedirs(args.ckpt_dir)
lr = args.lr * np.ones([args.epoch])
# reduce learning rate after 60% total epoch
small_lr_pos = int(args.epoch * 0.6)
lr[small_lr_pos:] = lr[0] * 0.1
if args.use_gpu:
# added to control the gpu memory
print("Run Tensorflow [GPU]\n")
gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=0.95)
with tf.Session(config=tf.ConfigProto(
gpu_options=gpu_options)) as sess:
model = denoiser(sess) # init a denoiser class
if args.phase == 'train':
denoiser_train(model, lr=lr)
elif args.phase == 'test':
# denoiser_test(model)
pass
else:
print('[!]Unknown phase')
sys.exit(1)
else:
# print("CPU\n")
print "CPU Not supported yet!"
sys.exit(1)
def test():
config = None
if args.use_gpu == 1:
config = tf.ConfigProto(gpu_options=tf.GPUOptions(
per_process_gpu_memory_fraction=0.9))
with tf.Session(config=config) as sess:
model = denoiser(sess)
denoiser_test(model)
def main(_):
if not os.path.exists(args.ckpt_dir):
os.makedirs(args.ckpt_dir)
if not os.path.exists(args.sample_dir):
os.makedirs(args.sample_dir)
if not os.path.exists(args.test_dir):
os.makedirs(args.test_dir)
#if not os.path.exists(args.save_dir):
# os.makedirs(args.save_dir)
if args.use_gpu:
# added to control the gpu memory
print("GPU\n")
# uses the GPU #0 on linux machines
os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
os.environ["CUDA_VISIBLE_DEVICES"] = "0"
print("\n\nExecution Environment:\n\n{}".format(
device_lib.list_local_devices()))
gpu_options = tf.compat.v1.GPUOptions(
per_process_gpu_memory_fraction=0.90)
with tf.compat.v1.Session(config=tf.compat.v1.ConfigProto(
gpu_options=gpu_options)) as sess:
model = denoiser(sess, args.batch_size)
if args.phase == 'train':
denoiser_train(model, lr=args.lr)
elif args.phase == 'test':
denoiser_test(model)
else:
print('[!]Unknown phase')
exit(0)
else:
print("CPU\n")
with tf.compat.v1.Session() as sess:
model = denoiser(sess, batch_size=args.batch_size)
if args.phase == 'train':
denoiser_train(model, lr=args.lr)
elif args.phase == 'test':
pdb.set_trace()
denoiser_test(model)
else:
print('[!]Unknown phase')
exit(0)
def main(_):
if args.use_gpu:
# added to control the gpu memory
print("Run Tensorflow [GPU]\n")
gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=0.9)
with tf.Session(config=tf.ConfigProto(gpu_options=gpu_options)) as sess:
model = denoiser(sess)
denoiser_test(model)
else:
print "CPU Not supported yet!"
sys.exit(1)
def main(_):
if not os.path.exists(args.ckpt_dir):
os.makedirs(args.ckpt_dir)
if not os.path.exists(args.sample_dir):
os.makedirs(args.sample_dir)
if not os.path.exists(args.test_dir):
os.makedirs(args.test_dir)
config = tf.ConfigProto(
device_count={"CPU": 32}, # limit to num_cpu_core CPU usage
inter_op_parallelism_threads=16,
intra_op_parallelism_threads=32)
lr = args.lr * np.ones([args.epoch])
lr[30:] = lr[0] / 10.0
if args.use_gpu:
# added to control the gpu memory
print("GPU\n")
gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=0.9)
with tf.Session(config=tf.ConfigProto(
gpu_options=gpu_options)) as sess:
model = denoiser(sess, sigma=args.sigma)
if args.phase == 'train':
denoiser_train(model, lr=lr)
elif args.phase == 'test':
denoiser_test(model)
else:
print('[!]Unknown phase')
exit(0)
else:
print("CPU\n")
with tf.device("/cpu:0"):
with tf.Session(config=config) as sess:
model = denoiser(sess, sigma=args.sigma)
if args.phase == 'train':
denoiser_train(model, lr=lr)
elif args.phase == 'test':
denoiser_test(model)
else:
print('[!]Unknown phase')
exit(0)
def main(_):
if not os.path.exists(args.ckpt_dir):
os.makedirs(args.ckpt_dir)
if not os.path.exists(args.sample_dir):
os.makedirs(args.sample_dir)
if not os.path.exists(args.test_dir):
os.makedirs(args.test_dir)
print(args)
lr_init = args.lr
lr_decay = [1.0, 0.2]
if args.use_gpu:
# added to control the gpu memory
print("GPU\n")
gpu_options = tf.GPUOptions(allow_growth=True)
#gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=0.7)
print(gpu_options)
with tf.Session(config=tf.ConfigProto(
gpu_options=gpu_options)) as sess:
model = denoiser(sess,
sigma=args.sigma,
kstage=args.kstage,
batch_size=args.batch_size,
config=args)
if args.phase == 'train':
denoiser_train(model, lr_init=lr_init, lr_decay=lr_decay)
elif args.phase == 'test':
denoiser_test(model)
else:
print('[!]Unknown phase')
exit(0)
else:
print("CPU\n")
with tf.Session() as sess:
model = denoiser(sess, sigma=args.sigma)
if args.phase == 'train':
denoiser_train(model, lr=lr)
elif args.phase == 'test':
denoiser_test(model)
else:
print('[!]Unknown phase')
exit(0)
def main(_):
if args.use_gpu:
# added to control the gpu memory
print("GPU\n")
gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=0.9)
with tf.Session(config=tf.ConfigProto(
gpu_options=gpu_options)) as sess:
model = denoiser(sess)
if args.phase == 'inference':
denoiser_inference(model)
else:
print('[!]Unknown phase')
exit(0)
else:
print("CPU\n")
with tf.Session() as sess:
model = denoiser(sess)
if args.phase == 'inference':
denoiser_inference(model)
else:
print('[!]Unknown phase')
exit(0)
def main(_):
if not os.path.exists(args.ckpt_dir):
os.makedirs(args.ckpt_dir)
if not os.path.exists(args.sample_dir):
os.makedirs(args.sample_dir)
if not os.path.exists(args.test_dir):
os.makedirs(args.test_dir)
lr = args.lr * np.ones([args.epoch])
lr[30:] = lr[0] / 10.0
if args.use_gpu:
# added to control the gpu memory
print("GPU\n")
#gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=0.9)
config = tf.ConfigProto(allow_soft_placement=True)
#config.gpu_options.allow_growth = True
#config.gpu_options.per_process_gpu_memory_fraction = 0.9
with tf.Session(config=config) as sess:
#with tf.device('/gpu:1'), tf.Session() as sess:
model = denoiser(sess, sigma=args.sigma)
if args.phase == 'train':
denoiser_train(model, lr=lr)
elif args.phase == 'test':
denoiser_test(model)
else:
print('[!]Unknown phase')
exit(0)
else:
print("CPU\n")
with tf.Session() as sess:
model = denoiser(sess, sigma=args.sigma)
if args.phase == 'train':
denoiser_train(model, lr=lr)
elif args.phase == 'test':
denoiser_test(model)
else:
print('[!]Unknown phase')
exit(0)
def main(_):
if not os.path.exists(checkpoint_dir):
os.makedirs(checkpoint_dir)
if not os.path.exists(sample_dir):
os.makedirs(sample_dir)
if not os.path.exists(test_dir):
os.makedirs(test_dir)
if args.use_gpu:
# added to control the gpu memory
print("GPU\n")
gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=0.9)
with tf.Session(config=tf.ConfigProto(
gpu_options=gpu_options)) as sess:
model = denoiser(sess, ip=ip)
if args.phase == 'train':
denoiser_train(model)
elif args.phase == 'test':
denoiser_test(model)
elif args.phase == 'test_repeatability':
denoiser_test2(model)
elif args.phase == 'generate_map':
denoiser_generate_map(model)
else:
print('[!]Unknown phase')
exit(0)
else:
print("CPU\n")
with tf.Session() as sess:
model = denoiser(sess, ip=ip)
if args.phase == 'train':
denoiser_train(model, lr=lr)
elif args.phase == 'test':
denoiser_test(model)
else:
print('[!]Unknown phase')
exit(0)
def main(_):
if not os.path.exists(args.ckpt_dir):
os.makedirs(args.ckpt_dir)
if not os.path.exists(args.sample_dir):
os.makedirs(args.sample_dir)
if not os.path.exists(args.test_dir):
os.makedirs(args.test_dir)
if not os.path.exists('./tmp_jpeg'):
os.makedirs('./tmp_jpeg')
lr = args.lr * np.ones([args.epoch])
lr[30:] = lr[0] / 10.0
if args.use_gpu:
# added to control the gpu memory
print("GPU\n")
gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=0.9)
with tf.Session(config=tf.ConfigProto(gpu_options=gpu_options)) as sess:
model = denoiser(sess, sigma=args.sigma)
if args.phase == 'train':
denoiser_train(model, lr=lr)
elif args.phase == 'test':
denoiser_test(model)
else:
print('[!]Unknown phase')
exit(0)
else:
print("CPU\n")
with tf.Session() as sess:
model = denoiser(sess, sigma=args.sigma)
if args.phase == 'train':
denoiser_train(model, lr=lr)
elif args.phase == 'test':
denoiser_test(model)
else:
print('[!]Unknown phase')
exit(0)
def main(_):
if not os.path.exists(args.ckpt_dir):
os.makedirs(args.ckpt_dir)
if not os.path.exists(args.test_dir):
os.makedirs(args.test_dir)
lr = args.lr * np.ones([args.epoch])
lr[30:] = lr[0] / 10.0
print("CPU\n")
with tf.Session() as sess:
model = denoiser(sess)
if args.phase == 'train':
denoiser_train(model, lr=lr)
elif args.phase == 'test':
denoiser_test(model)
else:
print('[!]Unknown phase')
exit(0)
def test_SAR2SAR(img):
sys.path.append('SAR2SAR-GRD-test')
import tensorflow.compat.v1 as tf
tf.disable_v2_behavior()
from model import denoiser
tf.reset_default_graph()
# DEFINE PARAMETERS OF SPECKLE AND NORMALIZATION FACTOR
M = 10.089038980848645
m = -1.429329123112601
L = 1
c = (1 / 2) * (special.psi(L) - np.log(L))
cn = c / (M - m) # normalized (0,1) mean of log speckle
def normalize_sar(im):
return ((np.log(im + np.spacing(1)) - m) * 255 / (M - m)).astype('float32')
img = normalize_sar(img)/255.0
with tf.Session() as sess:
model = denoiser(sess)
model.load('SAR2SAR-GRD-test/checkpoint')
Y_ = tf.placeholder(tf.float32, [None, None, None, 1],
name='clean_image')
pred = sess.run([model.Y], feed_dict={model.Y_: img})
sys.path.remove('SAR2SAR-GRD-test')
def denormalize_sar(im):
return np.exp((M - m) * (np.squeeze(im)).astype('float32') + m)
return pred[0][0,...]
def test():
# gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=0.9)
with tf.Session() as sess:
model = denoiser(sess)
denoiser_test(model)
def main(_):
if not os.path.exists(args.ckpt_dir):
os.makedirs(args.ckpt_dir)
if not os.path.exists(args.sample_dir):
os.makedirs(args.sample_dir)
if not os.path.exists(args.test_dir):
os.makedirs(args.test_dir)
config = tf.ConfigProto(
device_count={"CPU": 32}, # limit to num_cpu_core CPU usage
inter_op_parallelism_threads=16,
intra_op_parallelism_threads=32)
lr = args.lr * np.ones([args.epoch])
lr[30:] = lr[0] / 10.0
if args.phase == 'test':
with tf.device("/cpu:0"):
with tf.Session(config=config) as sess:
model = denoiser(sess, sigma=args.sigma, isDenoise=True)
denoiser_test(model)
elif args.phase == 'testcmp':
with tf.device("/cpu:0"):
with tf.Session(config=config) as sess:
model = cmpdenoiser(sess)
denoiser_test(model)
elif args.phase == 'train':
# distribution check
server, cluster = createServer()
if FLAGS.job_name == "ps":
server.join()
elif FLAGS.job_name == "worker":
worker_device = "/job:worker/task:%d" % FLAGS.task_index
with tf.device(
tf.train.replica_device_setter(worker_device=worker_device,
ps_device="/job:ps/cpu:0",
cluster=cluster)):
model = denoiser(sigma=args.sigma)
denoiser_train(model, server, FLAGS.task_index, lr=lr)
elif args.phase == 'trainconv':
server, cluster = createServer()
if FLAGS.job_name == "ps":
server.join()
elif FLAGS.job_name == "worker":
worker_device = "/job:worker/task:%d" % FLAGS.task_index
with tf.device(
tf.train.replica_device_setter(worker_device=worker_device,
ps_device="/job:ps/cpu:0",
cluster=cluster)):
model = convdenoiser()
conv_denoise_train(model, server, FLAGS.task_index, lr=lr)
elif args.phase == 'trainconvpatch':
server, cluster = createServer()
if FLAGS.job_name == "ps":
server.join()
elif FLAGS.job_name == "worker":
worker_device = "/job:worker/task:%d" % FLAGS.task_index
with tf.device(
tf.train.replica_device_setter(worker_device=worker_device,
ps_device="/job:ps/cpu:0",
cluster=cluster)):
model = convpatchdenoiser(num_workers=len(
FLAGS.worker_hosts.split(",")),
is_chief=FLAGS.task_index == 0)
conv_patch_denoise_train(model, server, FLAGS.task_index)
elif args.phase == 'traincmp':
server, cluster = createServer()
if FLAGS.job_name == "ps":
server.join()
elif FLAGS.job_name == "worker":
worker_device = "/job:worker/task:%d" % FLAGS.task_index
with tf.device(
tf.train.replica_device_setter(worker_device=worker_device,
ps_device="/job:ps/cpu:0",
cluster=cluster)):
cmp_denoise_train(server,
FLAGS.task_index,
num_worker=FLAGS.worker_hosts.split(","))
elif args.phase == 'compare':
denoise_files = glob('./data/test/{}/*.jpg'.format(args.denoise_set))
noise_files = glob('./data/test/{}/*.jpg'.format(args.denoise_set +
'_nodenoise'))
diffs = 0.0
for file in denoise_files:
noise_file = find_match_file([file], noise_files)
noise_image = load_images(noise_file[0]).astype(np.float32) / 255.0
denoise_image = load_images(file).astype(np.float32) / 255.0
save_img = np.clip(255 * noise_image, 0, 255).astype('uint8')
save_images(
os.path.join(args.test_dir,
file.split("/")[-1].split("_")[0] + ".jpg"),
save_img)
diff = cal_psnr(denoise_image, noise_image)
diffs += diff
diffs = diffs / len(denoise_files)
elif args.phase == 'alone_conv_patch':
with tf.device("/cpu:0"):
model = convpatchdenoiser()
conv_patch_denoise_train(model)
elif args.phase == 'batch_cmp':
place = [n for n in range(0, 20)]
print(place)
np.random.shuffle(place)
print(place)
# with load_data(filepath='./data/img_denoise_pats.npy', rand=False) as data_denoise:
# with load_data(filepath='./data/img_noise_pats.npy', rand=False) as data_noise:
# numBatch = int(data_denoise.shape[0] / 128)
# noise_num = int(data_denoise.shape[0] / 128)
# print ('%d---%d' % (numBatch, noise_num))
# for i in range(0, 10):
# num = random.randint(0,numBatch)
# print (num)
# denoise = data_denoise[num:num + 1, :, :, :]
# noise = data_noise[num:num + 1, :, :, :]
# save_images(os.path.join(args.test_dir, 'denoised%d.jpg' % i), denoise)
# save_images(os.path.join(args.test_dir, 'noise%d.jpg' % i), noise)
else:
print('[!]Unknown phase')
exit(0)
def main(_):
#the following string is attached to checkpoint, log and image folder names
eps_str = str(int(np.log10(args.eps)))
name = "NET_" + args.cost + "_eps1e" + eps_str + "_sigma" + str(
int(args.sigma)) + "_" + str(args.type)
ckpt_dir = args.ckpt_dir + "/" + name
sample_dir = args.sample_dir + "/" + name
test_dir = args.test_dir + "/" + name
log_dir = args.log_dir + "/" + name
if not os.path.exists(ckpt_dir):
os.makedirs(ckpt_dir)
#learning rate decay schedule
lr = args.lr * np.ones([args.epoch])
lr[100:] = lr[0] / 10.0 #lr decay after 100 epochs
#dataset
mnist = input_data.read_data_sets('MNIST_data', one_hot=True)
print("Model: %s" % (name))
if args.use_gpu:
# added to control the gpu memory
print("GPU\n")
gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=0.9)
with tf.Session(config=tf.ConfigProto(
gpu_options=gpu_options)) as sess:
model = denoiser(sess,
dataset=mnist,
sigma=args.sigma,
eps=args.eps,
cost_str=args.cost,
ckpt_dir=ckpt_dir,
sample_dir=sample_dir,
log_dir=log_dir)
if args.phase == 'train':
denoiser_train(model, lr=lr)
elif args.phase == 'test':
denoiser_test(model, test_dir)
else:
print('[!]Unknown phase')
exit(0)
else:
print("CPU\n")
with tf.Session() as sess:
model = denoiser(sess,
dataset=mnist,
sigma=args.sigma,
eps=args.eps,
cost_str=args.cost,
ckpt_dir=ckpt_dir,
sample_dir=sample_dir,
log_dir=log_dir)
if args.phase == 'train':
denoiser_train(model, lr=lr)
elif args.phase == 'test':
denoiser_test(model, test_dir)
else:
print('[!]Unknown phase')
exit(0)
checkpoint_dir = '/content/SAR2SAR-test/checkpoint'
if not os.path.exists(args.test_dir):
os.makedirs(args.test_dir)
from model import denoiser
def denoiser_test(denoiser):
test_data = args.test_data
print(
"[*] Start testing on real data. Working directory: %s. Collecting data from %s and storing test results in %s"
% (os.getcwd(), test_data, args.test_dir))
test_files = glob((test_data + '/*.npy').format('float32'))
denoiser.test(test_files,
ckpt_dir=checkpoint_dir,
save_dir=args.test_dir,
dataset_dir=test_data,
stride=args.stride_size)
if __name__ == '__main__':
if args.use_gpu:
gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=0.9)
with tf.Session(config=tf.ConfigProto(
gpu_options=gpu_options)) as sess:
model = denoiser(sess)
denoiser_test(model)
else:
with tf.Session() as sess:
model = denoiser(sess)
denoiser_test(model)
# encoding: utf-8
from options import Options
from model import denoiser
from data_loader import load_test_data
import time
if __name__ == '__main__':
args = Options().initialize()
model = denoiser(args)
if args.eval:
model.eval()
# data = load_test_data(args)
# model.set_test_input(data)
model.test_new(args, 64)
print(res)
file.write(res + '\n')
#---------------------------------------------------------------------------
# experiment 2 : We denoise this image using a CNN Gaussian Denoiser(train with quantile) with all sigma
# Goal see the influence of the CNN
#---------------------------------------------------------------------------
list_ckp = [
"./checkpoint", "./checkpoint_sar", "checkpoint_sar_norm",
"./checkpoint_sar_1"
]
list_name = ["DuCNN25Nat", "DuCNN25Sar_quant", "DuCNN25Sar", "DuCNN13SarQuant"]
list_lamb = [1, 1, 1, 3]
if (True):
print("eperiment 2:")
model = denoiser(sess, sigma=25, add_noise=False)
list_beta = [1, 1, 1, 4]
for i in range(len(list_beta)):
ckpt_dir = list_ckp[i]
model.load(ckpt_dir)
C = np.max(img_in) - np.min(img_in)
b = img_in.min() / C
img_inp = img_in / C - b
img_input = img_inp.reshape(1, img_in.shape[1], img_in.shape[0], 1)
out2, _, _ = model.denoise(img_input)
out2 = out2.reshape(img_in.shape)
out2 = C * (out2 + b)
out2 = out2 + utils.debiased(args.L)
def test():
gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=0.9)
with tf.Session(config=tf.ConfigProto(gpu_options=gpu_options)) as sess:
model = denoiser(sess)
denoiser_test(model)
# encoding: utf-8
from options import Options
from model import denoiser
from data_loader import Data_loader
import time
from torch.utils.data import DataLoader
import torchvision.utils as tv
from tensorboardX import SummaryWriter
import numpy as np
if __name__ == '__main__':
args=Options().initialize()
model = denoiser(args)
writer = SummaryWriter('runs/threshold_16_vflip')
start_epoch = 0
start_time = time.time()
print("start training....")
model.print_networks(args.verbose)
niter=0
dataset = Data_loader()
dataloader = DataLoader(dataset, batch_size=32, shuffle=True, num_workers=0, drop_last=False)
batch_num=len(dataloader)
for epoch in range(start_epoch, args.epoch):
batch_id=0
for labels, inputs in dataloader:
batch_id = batch_id + 1
data=[labels, inputs]
input_widefield,input_confocal= model.set_input(data)
# model.evaluate(iter_num, args) # whether pretrained
model.optimize_parameters()
loss = model.get_loss()
