def __init__(self, args):
super(Model, self).__init__()
self.se_resnet = se_resnet18()
self.convnet = convnet()
self.fc_in = torch.nn.Linear(256, 256)
self.fc_out = torch.nn.Linear(256, 199)
def fit_convnet(cell, stimulus_type):
"""
Demo code for fitting a convnet model
"""
# initialize model
mdl = convnet(cell, stimulus_type, num_filters=(8, 16), filter_size=(13, 13),
weight_init='normal', l2_reg=0.01, mean_adapt=False)
# train
batchsize = 5000 # number of samples per batch
num_epochs = 10 # number of epochs to train for
save_weights_every = 50 # save weights every n iterations
mdl.train(batchsize, num_epochs=num_epochs, save_every=save_weights_every)
return mdl
def _build_model(self):
self.n_color_cls = 8
self.net = models.convnet(num_classes=self.option.n_class)
self.pred_net_r = models.Predictor(input_ch=32, num_classes=self.n_color_cls)
self.pred_net_g = models.Predictor(input_ch=32, num_classes=self.n_color_cls)
self.pred_net_b = models.Predictor(input_ch=32, num_classes=self.n_color_cls)
self.loss = nn.CrossEntropyLoss(ignore_index=255)
self.color_loss = nn.CrossEntropyLoss(ignore_index=255)
if self.option.cuda:
self.net.cuda()
self.pred_net_r.cuda()
self.pred_net_g.cuda()
self.pred_net_b.cuda()
self.loss.cuda()
self.color_loss.cuda()
def fit_convnet(cell, stimulus_type):
"""
Demo code for fitting a convnet model
"""
# initialize model
mdl = convnet(cell,
stimulus_type,
num_filters=(8, 16),
filter_size=(13, 13),
weight_init='normal',
l2_reg=0.01,
mean_adapt=False)
# train
batchsize = 5000 # number of samples per batch
num_epochs = 10 # number of epochs to train for
save_weights_every = 50 # save weights every n iterations
mdl.train(batchsize, num_epochs=num_epochs, save_every=save_weights_every)
return mdl
if __name__ == '__main__':
# Let's load and process the dataset
import numpy as np
from fuel.datasets.dogs_vs_cats import DogsVsCats
from fuel.streams import DataStream
from fuel.schemes import ShuffledScheme
from fuel.transformers.image import RandomFixedSizeCrop
from fuel.transformers import Flatten
# Load the training set
train = DogsVsCats(('train',),subset=slice(0, 20)) #subset=slice(0, 20000)
test = DogsVsCats(('test',),subset=slice(0,20))
input_size = (150,150)
from models import mlp,convnet
#main(None,mlp(input_size[0]*input_size[1]*3), train, test, num_epochs=1, input_size=input_size, batch_size=5, num_batches=20, flatten_stream=True)
main("test1.txt", convnet(input_size), train, test, num_epochs=1, input_size=input_size, batch_size=64, num_batches=100)
# from deep_res import build_cnn
# model = build_cnn(x,3,64)
#
# THEANO_FLAGS='cuda.root=/usr/lib/nvidia-cuda-toolkit/', THEANO_FLAGS=cuda.root=/usr/lib/nvidia-cuda-toolkit/,device=gpu,floatX=float32 python dogs_cats.py
# THEANO_FLAGS=device=gpu
# THEANO_FLAGS=exception_verbosity=high,optimizer=None
test_data = QD_Dataset(mtype="test", root=args.data_root)
test_loader = torch.utils.data.DataLoader(test_data,
batch_size=args.test_bs,
shuffle=True)
num_classes = train_data.get_number_classes()
print("Train images number: %d" % len(train_data))
print("Test images number: %d" % len(test_data))
net = None
if args.model == 'resnet34':
net = resnet34(num_classes)
elif args.model == 'convnet':
net = convnet(num_classes)
if args.ngpu > 1:
net = nn.DataParallel(net)
if args.ngpu > 0:
net.cuda()
print(net)
optimizer = torch.optim.SGD(net.parameters(),
state['learning_rate'],
momentum=state['momentum'],
weight_decay=state['weight_decay'])
Train_Loss = []
Img[i][j] = [1, 1, 1]
else:
Img[i][j] = [
1 - canvas[i * shape + j] / 255,
1 - canvas[i * shape + j] / 255,
1 - canvas[i * shape + j] / 255
]
plt.imshow(Img)
plt.title(name)
plt.savefig("./" + root + "/ans/" + str(idx) + "_" + name + ".png")
net = None
if args.model == 'resnet34':
net = resnet34(args.num_classes)
elif args.model == 'convnet':
net = convnet(args.num_classes)
if args.ngpu > 1:
net = nn.DataParallel(net)
if args.ngpu > 0:
net.cuda()
net.load_state_dict(torch.load('./' + args.model_file_name))
print("Model loaded, start evaluating.")
generate_eval_dataset()
eval_data = QD_Dataset(mtype="eval", root="Dataset")
eval_loader = torch.utils.data.DataLoader(eval_data,
batch_size=args.eval_bs,
if __name__ == '__main__':
print('PAC 2018')
example = data_dir + subjects[0]['id'] + '.nii'
image_size = nib.load(example).shape
print('Image size:', image_size)
features = np.zeros((len(subjects), 4))
labels = np.zeros((len(subjects), 2))
for i, sub in enumerate(subjects):
features[i, 0] = sub['tiv']
features[i, 1] = sub['site']
features[i, 2] = sub['gender']
features[i, 3] = sub['age']
if sub['label'] == 1:
labels[i, :] = [1, 0]
else:
labels[i, :] = [0, 1]
f = hdf5_smash(subjects)
model = convnet(image_size)
model.fit_generator(batch_gen(f, labels, features),
steps_per_epoch=None,
epochs=10,
validation_data=0.1,
shuffle=False)