build_model=build_model,
optimizer=optimizer,
loss=loss,
lr_sub_size=param['lr_sub_size'],
lr_sub_stride=param['lr_sub_stride'],
random=random,
save_dir=Path('./results') / save_dir)
if args.export_only:
expt.export_pb_model(['input_lr'], ['output_hr'],
Path('./results') / save_dir / 'model.pb', export_bgr)
exit()
# Training
expt.train(train_set=param['train_set'],
val_set=param['val_set'],
epochs=param['epochs'],
batch_size=param['batch'],
resume=True)
# Evaluation
if 'test_sets' in param:
for test_set in param['test_sets']:
expt.test(test_set=test_set)
if 'test_files' in param:
expt.test_file(param['test_files'])
# Export tensorflow .pb model
expt.export_pb_model(['input_lr'], ['output_hr'],
Path('./results') / save_dir / 'model.pb', export_bgr)
parser = argparse.ArgumentParser()
parser.add_argument('param_file', type=Path)
args = parser.parse_args()
param = json.load(args.param_file.open())
# Model
scale = param['scale']
build_model = partial(get_model(param['model']['name']),
**param['model']['params'])
if 'optimizer' in param:
optimizer = getattr(optimizers, param['optimizer']['name'].lower())
optimizer = optimizer(**param['optimizer']['params'])
else:
optimizer = 'adam'
# Data
load_set = partial(load_set,
lr_sub_size=param['lr_sub_size'],
lr_sub_stride=param['lr_sub_stride'])
# Training
expt = Experiment(scale=param['scale'], load_set=load_set,
build_model=build_model,optimizer=optimizer,
save_dir=param['save_dir'])
expt.train(train_set=param['train_set'], val_set=param['val_set'],
epochs=param['epochs'], resume=True)
# Evaluation
for test_set in param['test_sets']:
expt.test(test_set=test_set)
from toolbox.image import bicubic_resize
# Model
scale = 3
model = espcn(c=1, f1=9, f2=1, f3=5, n1=64, n2=32)
model.summary()
# Data
train_set = '91-image'
val_set = 'Set5'
test_sets = ['Set5', 'Set14']
preprocess = partial(bicubic_resize, size=scale)
load_set = partial(load_set,
sub_size=11,
sub_stride=5,
scale=scale,
channel=0,
preprocess=preprocess)
# Training
experiment = Experiment(scale=scale,
model=model,
preprocess=preprocess,
load_set=load_set,
save_dir='.')
experiment.train(train_set=train_set, val_set=val_set, epochs=500, resume=True)
# Evaluation
for test_set in test_sets:
experiment.test(test_set=test_set)
**param['model']['params'])
if 'optimizer' in param:
optimizer = getattr(optimizers, param['optimizer']['name'].lower())
optimizer = optimizer(**param['optimizer']['params'])
else:
optimizer = 'adam'
lr_block_size = tuple(param['lr_block_size'])
# Data
load_train_set = partial(load_train_set,
lr_sub_size=param['lr_sub_size'],
lr_sub_stride=param['lr_sub_stride'])
# Training
expt = Experiment(scale=param['scale'],
load_set=load_train_set,
build_model=build_model,
optimizer=optimizer,
save_dir=param['save_dir'])
print('training process...')
expt.train(train_set=param['train_set'],
val_set=param['val_set'],
epochs=param['epochs'],
resume=True)
# Evaluation
print('evaluation process...')
for test_set in param['test_sets']:
expt.test(test_set=test_set, lr_block_size=lr_block_size)