refDir = valR[i].replace("R_no_atoms", "A_no_atoms")
Y = np.array(np.array(
io.imread(
valR[i])[int(inL / 2 - outL / 2):int(inL / 2 + outL / 2),
int(inL / 2 - outL / 2):int(inL / 2 + outL / 2)]),
dtype=np.dtype('float32')) / 4294967295
ref = np.array(np.array(
io.imread(refDir)[int(inL / 2 - outL / 2):int(inL / 2 +
outL / 2),
int(inL / 2 - outL / 2):int(inL / 2 +
outL / 2)]),
dtype=np.dtype('float32')) / 4294967295
valRloss[i] = np.average((ref - Y)**2)
np.save('referenceMSEvalR.npy', valRloss)
referenceMSE = np.mean(
np.concatenate((trainAloss, trainRloss, valAloss, valRloss)))
print('reference MSE is ' + str(referenceMSE))
return referenceMSE
if __name__ == '__main__':
from unet import unet_model
model = unet_model(476, 192)
model = initialize_model(model)
from preprocessing import prepare_datasets
trainList, valList, testList = prepare_datasets(476, 442, 804, 0.2)
referenceMSE = generate_referance_loss(476, 192, trainList, valList)
valR = [s for s in valList if "R_no_atoms" in s]
valRloss = np.empty(len(valR))
for i in tqdm(range(len(valR))):
refDir = valR[i].replace("R_no_atoms", "A_no_atoms")
Y = np.array(np.array(io.imread(valR[i])[int(inL / 2 - outL / 2):int(inL / 2 + outL / 2),
int(inL / 2 - outL / 2):int(inL / 2 + outL / 2)]),
dtype=np.dtype('float32')) / 4294967295
ref = np.array(np.array(io.imread(refDir)[int(inL / 2 - outL / 2):int(inL / 2 + outL / 2),
int(inL / 2 - outL / 2):int(inL / 2 + outL / 2)]),
dtype=np.dtype('float32')) / 4294967295
valRloss[i] = np.average((ref - Y) ** 2)
np.save('referenceMSEvalR.npy', valRloss)
referenceMSE = np.mean(np.concatenate((trainAloss, trainRloss, valAloss, valRloss)))
print('reference MSE is ' + str(referenceMSE))
return referenceMSE
if __name__ == '__main__':
from unet import unet_model
model = unet_model(476, 192)
model = initialize_model(model)
from preprocessing import prepare_datasets
# trainList, valList, testList = prepare_datasets(476, 442, 804, 0.2)
trainList, valList, testList = prepare_datasets(804, 0.2)
referenceMSE = generate_reference_loss(476, 192, trainList, valList)
'use this flag if you don\'t have the same dataset structure as the original: '
'A_no_atoms, R_no_atoms, A_with_atoms, R_with_atoms')
return parser
if __name__ == '__main__':
## params
parser = get_parser()
args = parser.parse_args()
outL = 2 * args.maskR # Output size
mask = generate_mask(args.inL, args.maskR)
## get data
trainList, valList, testList = prepare_datasets(args.inL, args.centerVer,
args.centerHor, 0.2)
## build model
K.clear_session()
model = unet_model(args.inL, outL)
model.summary() # display model summary
model, epochNum, trainLoss, valLoss = initialize_model(model)
if args.SGD:
opt = SGD(lr=1e-2, momentum=0.9, decay=1e-4 / args.max_epochs)
else:
opt = Adam(lr=args.learning_rate)
#model.compile(optimizer=opt, loss='mse')
model.compile(optimizer=opt, loss='mse')
## calculate referance loss
'use this flag if you don\'t have the same dataset structure as the original: '
'A_no_atoms, R_no_atoms, A_with_atoms, R_with_atoms')
return parser
if __name__ == '__main__':
## params
parser = get_parser()
args = parser.parse_args()
outL = 2 * args.maskR # Output size
mask = generate_mask(args.inL, args.maskR)
## get data
trainList, valList, testList = prepare_datasets(args, 0.2)
## build model
K.clear_session()
model = unet_model(args.inL, outL)
model.summary() # display model summary
model, epochNum, trainLoss, valLoss = initialize_model(model)
if args.SGD:
opt = SGD(lr=1e-2, momentum=0.9, decay=1e-4 / args.max_epochs)
else:
opt = Adam(lr=args.learning_rate)
model.compile(optimizer=opt, loss='mse')
model.compile(optimizer=opt, loss='mse')
## calculate referance loss
if not args.skip_reference_comparison: