def test_iris(self):
scores = []
for i in range(1):
hidden = 50
l1 = Linear(4, hidden, initialize='ones')
l2 = Linear(hidden, 3, initialize='ones')
l1.W *= 0.000000001
l2.W *= 0.00000001
model = Seq(l1, Sigmoid, l2)
loss = CrossEntropyLoss()
trainer = OnlineTrainer()
losses = trainer.train(model,
self.train_set,
epochs=100,
loss=loss,
optimizer=SGD(learning_rate=0.01))
score = loss.test_score(model, self.train_set)
print("hidden=%f score=%f" % (hidden, score))
scores.append(score)
self.plot_loss_history(losses)
plt.show()
self.assertGreaterEqual(numpy.mean(scores), 94.)
def network_setup(model_file_path=None):
freq_count = 4000
count_bins = 88 * 20
dataset = MapsDB('../db',
freq_count=freq_count,
count_bins=count_bins,
batch_size=128,
start_time=0.5,
duration=0.5)
model = Network()
model.add(Linear('fc1', dataset.get_vec_input_width(), 2048, 0.001))
model.add(Sigmoid('sigmoid1'))
model.add(Linear('fc2', 2048, dataset.get_label_width(), 0.001))
model.add(Softmax('softmax2'))
loss = CrossEntropyLoss(name='xent')
# loss = EuclideanLoss(name='r2')
optim = SGDOptimizer(learning_rate=0.00001, weight_decay=0.005, momentum=0.9)
# optim = AdagradOptimizer(learning_rate=0.001, eps=1e-6)
input_placeholder = T.fmatrix('input')
label_placeholder = T.fmatrix('label')
label_active_size_placeholder = T.ivector('label_active_size')
if model_file_path:
model.loads(model_file_path)
else:
dataset.load_cache()
model.compile(input_placeholder, label_placeholder, label_active_size_placeholder, loss, optim)
return model, dataset, freq_count, count_bins
def set_model():
model = WideResnet(n_classes, k=wresnet_k, n=wresnet_n) # wide resnet-28
model.train()
model.cuda()
criteria_x = CrossEntropyLoss().cuda()
criteria_u = nn.MSELoss().cuda()
return model, criteria_x, criteria_u
def getNetwork():
'''
to obtain network structure from specified file
'''
file_name = "models/structure.json"
if len(sys.argv)>1:
file_name = sys.argv[1]
f = file(file_name, "r")
s = f.read()
f.close()
networks = json.loads(s)
for network in networks:
config = network['config']
dis_model = network['model']
model = Network()
for layer in dis_model:
if layer['type'] == 'Linear':
model.add(Linear(layer['name'], layer['in_num'], layer['out_num'], layer['std']))
if layer['type'] == 'Relu':
model.add(Relu(layer['name']))
if layer['type'] == 'Sigmoid':
model.add(Sigmoid(layer['name']))
if layer['type'] == 'Softmax':
model.add(Softmax(layer['name']))
loss = EuclideanLoss('loss')
if 'loss' in config:
if config['loss'] == 'CrossEntropyLoss':
loss = CrossEntropyLoss('loss')
yield network['name'], model, config, loss
def run(self,
batch_size=10,
learning_rate=0.6,
train_set_percentage=1.0,
epochs=3):
model = Seq(
WxBiasLinear(784, 10, initialize='random')
# Dropout(0.5)
)
train_set_sliced = slice_percentage(self.train_set,
train_set_percentage)
trainer = MinibatchTrainer()
trainer.train_minibatches(
model,
train_set_sliced,
batch_size=batch_size,
loss=CrossEntropyLoss(),
epochs=epochs,
optimizer=SGD(learning_rate=learning_rate),
# optimizer=RMSProp(learning_rate=learning_rate, decay_rate=0.6),
# optimizer=AdaGrad(learning_rate=learning_rate),
show_progress=True)
# self.show_mnist_grid(model, self.test_set)
# trainer = PatienceTrainer()
# trainer.train(model,
# train_set_sliced, self.valid_set, self.test_set,
# batch_size=batch_size,
# loss=CrossEntropyLoss(),
# max_epochs=100,
# # optimizer=MomentumSGD(learning_rate=learning_rate, momentum=0.5),
# optimizer=RMSProp(learning_rate=learning_rate, decay_rate=0.9),
# # optimizer=AdaGrad(learning_rate=learning_rate),
# test_score_function=self.test_score_fun
# )
test_score = CrossEntropyLoss().test_score(model, self.test_set)
return {
# 'train_score': train_score,
'test_score': test_score,
}
def __init__(self, args):
super(MultiTaskNet, self).__init__()
self.encoder = DGCNN_Seg_Encoder(args)
self.decoder = FoldNet_Decoder(args)
if args.rec_loss == 'ChamferLoss':
self.rec_loss = ChamferLoss()
elif args.rec_loss == 'ChamferLoss_m':
self.rec_loss = ChamferLoss_m()
self.classifer = DGCNN_RT_Cls_Classifier(args)
self.rot_loss = CrossEntropyLoss()
def __init__(self, args, seg_num_all=50):
super(SegmentationNet, self).__init__()
self.is_eval = args.eval
if args.encoder == 'foldnet':
self.encoder = FoldNet_Encoder(args)
elif args.encoder == 'dgcnn_cls':
self.encoder = DGCNN_Cls_Encoder(args)
elif args.encoder == 'dgcnn_seg':
self.encoder = DGCNN_Seg_Encoder(args)
if not self.is_eval:
self.segmenter = DGCNN_Seg_Segmenter(args, seg_num_all)
self.loss = CrossEntropyLoss()
def __init__(self, args):
super(ClassificationNet, self).__init__()
self.is_eval = args.eval
if args.encoder == 'foldnet':
self.encoder = FoldNet_Encoder(args)
elif args.encoder == 'dgcnn_cls':
self.encoder = DGCNN_Cls_Encoder(args)
elif args.encoder == 'dgcnn_seg':
self.encoder = DGCNN_Seg_Encoder(args)
if not self.is_eval:
self.classifier = DGCNN_Cls_Classifier(args)
self.loss = CrossEntropyLoss()
def run(self):
train_set, valid_set, test_set = gen_data(dataset='circles')
l1 = 0.
l2 = 0.
# model = Seq([
# Linear(2, 100),
# # Dropout(0.8),
# Tanh(),
# Linear(100, 4),
# # Tanh(),
# # Dropout(0.6)
# ])
model = SyntaxLayer(
syntax.Linear(10, 4, input=
syntax.Tanh(syntax.Linear(2, 10, input=syntax.Var('x')))))
# trainer = SimpleTrainer()
# trainer.train(model, train_set,
# loss=CrossEntropyLoss(),
# optimizer=SGD(learning_rate=0.1),
# # optimizer=MomentumSGD(learning_rate=0.1, momentum=0.8),
# # optimizer=AdaGrad(learning_rate=0.9),
# # optimizer=RMSProp(learning_rate=0.1, decay_rate=0.9),
# epochs=100)
trainer = MinibatchTrainer()
batch_size = 40
batches_n = (float(len(train_set)) / batch_size)
learning_rate = 0.1 / batches_n
# print learning_rate
mean_losses = trainer.train_minibatches(model, train_set,
batch_size,
epochs=100,
loss=CrossEntropyLoss(),
optimizer=MomentumSGD(learning_rate, momentum=0.4),
show_progress=True)
# trainer = PatienceTrainer()
# mean_losses = trainer.train(model, train_set, valid_set, test_set,
# batch_size=1,
# max_epochs=10000,
# loss=CrossEntropyLoss(),
# test_score_function=CrossEntropyLoss.test_score,
# optimizer=SGD(learning_rate=0.1))
draw_decision_surface(model)
scatter_train_data(train_set)
plot_mean_loss(mean_losses)
plt.show()
def __init__(self, args, seg_num_all=50):
super(SegmentationNet, self).__init__()
self.is_eval = args.eval
self.loss_type = args.loss
if args.encoder == 'foldnet':
self.encoder = FoldNet_Encoder(args)
elif args.encoder == 'dgcnn_cls':
self.encoder = DGCNN_Cls_Encoder(args)
elif args.encoder == 'dgcnn_seg':
self.encoder = DGCNN_Seg_Encoder(args)
if not self.is_eval:
self.segmenter = DGCNN_Seg_Segmenter(args, seg_num_all)
if self.loss_type == 'softmax':
self.loss = CrossEntropyLoss()
elif self.loss_type == 'triplet':
self.loss = TripletLoss(margin=args.margin)
def test_numerical_gradient(self):
x = np.random.rand(5)
target_class = make_one_hot_target(classes_n=5, target_class=1)
loss = CrossEntropyLoss()
y = loss.calc_loss(x, target_class)
grad = loss.calc_gradient(y, target_class)
def forward(i):
return loss.calc_loss(i, target_class)
num_grad = numerical_gradient.calc(forward, x)
num_grad = np.sum(num_grad, axis=0)
print num_grad
numerical_gradient.assert_are_similar(grad, num_grad)
def test_ClaudioMaxNLL(self):
x = np.random.rand(10)
y = np.random.rand(10)
target_class = 3
loss = CrossEntropyLoss()
def f1():
J = loss.calc_loss(x, target_class)
dJdy = loss.calc_gradient(y, target_class)
cla_loss = ClaudioMaxNLL()
def f2():
J = cla_loss.calc_loss(x, target_class)
dJdy = cla_loss.calc_gradient(y, target_class)
t1 = timeit.timeit(f1, number=10000)
t2 = timeit.timeit(f1, number=10000)
print(t1, t2)
self.assertGreater(t2, t1)
def train(seq, dataloader, epochs=10):
criterion = CrossEntropyLoss(seq)
optimizer = Adam(seq)
for epoch in range(epochs):
epoch_loss = 0.0
epoch_accuracy = 0.0
n_batch = 0
for batch, labels in dataloader:
n_batch += 1
outputs = seq(batch)
loss = criterion(outputs, labels)
accuracy = accuracy_score(outputs.argmax(axis=1), labels)
loss.backward()
optimizer.step()
epoch_loss += loss
epoch_accuracy += accuracy
print("Epoch {}/{} - loss: {:%.5f} accuracy: {:%.5f}".format(
epoch + 1, epochs, epoch_loss / n_batch, epoch_accuracy / n_batch))
print("Finished training !")
train_data, test_data, train_label, test_label = load_mnist_for_cnn('data')
model = Network()
model.add(Convolution('conv1', 5, 1, 8, 0.1, 28,
batch_size)) # output size: N x 4 x 24 x 24
model.add(Relu('relu1'))
model.add(Pooling('pool1', 2)) # output size: N x 4 x 12 x 12
model.add(Convolution('conv2', 5, 8, 16, 0.1, 12,
batch_size)) # output size: N x 8 x 10 x 10
model.add(Relu('relu2'))
model.add(Pooling('pool2', 2)) # output size: N x 8 x 5 x 5
model.add(Flatten('View')) # my own layer, to reshape the output
model.add(Linear('fc3', 16 * 4 * 4, 10,
0.1)) # input reshaped to N x 200 in Linear layer
model.add(Softmax('softmax'))
loss = CrossEntropyLoss(name='xent')
optim = AdagradOptimizer(learning_rate=0.0001, eps=1e-8)
input_placeholder = T.ftensor4('input')
label_placeholder = T.fmatrix('label')
model.compile(input_placeholder, label_placeholder, loss, optim)
print 'ada'
solve_net(model,
train_data,
train_label,
test_data,
test_label,
batch_size=batch_size,
max_epoch=50,
disp_freq=1000,
def main():
#GENERAL
torch.cuda.empty_cache()
root = "/home/kuru/Desktop/veri-gms-master/"
train_dir = '/home/kuru/Desktop/veri-gms-master/VeRispan/image_train/'
source = {'verispan'}
target = {'verispan'}
workers = 4
height = 320
width = 320
train_sampler = 'RandomSampler'
#AUGMENTATION
random_erase = True
jitter = True
aug = True
#OPTIMIZATION
opt = 'adam'
lr = 0.001
weight_decay = 5e-4
momentum = 0.9
sgd_damp = 0.0
nesterov = True
warmup_factor = 0.01
warmup_method = 'linear'
STEPS = (30, 60)
GAMMA = 0.1
WARMUP_FACTOR = 0.01
WARMUP_EPOCHS = 10
WARMUP_METHOD = 'linear'
#HYPERPARAMETER
max_epoch = 80
start = 0
train_batch_size = 16
test_batch_size = 50
#SCHEDULER
lr_scheduler = 'multi_step'
stepsize = [30, 60]
gamma = 0.1
#LOSS
margin = 0.3
num_instances = 4
lambda_tri = 1
#MODEL
#arch = 'resnet101'
arch = 'resnet50_ibn_a'
no_pretrained = False
#TEST SETTINGS
#load_weights = '/home/kuru/Desktop/veri-gms-master/IBN-Net_pytorch0.4.1/resnet101_ibn_a.pth'
#load_weights = '/home/kuru/Desktop/veri-gms-master/IBN-Net_pytorch0.4.1/resnet101_ibn_a.pth'
load_weights = '/home/kuru/Desktop/veri-gms-master/IBN-Net_pytorch0.4.1/resnet50_ibn_a.pth'
#load_weights = None
start_eval = 0
eval_freq = -1
num_classes = 776
feat_dim = 2048
CENTER_LR = 0.5
CENTER_LOSS_WEIGHT = 0.0005
center_criterion = CenterLoss(num_classes=num_classes,
feat_dim=feat_dim,
use_gpu=True)
optimizer_center = torch.optim.SGD(center_criterion.parameters(),
lr=CENTER_LR)
#MISC
use_gpu = True
#use_gpu = False
print_freq = 10
seed = 1
resume = ''
save_dir = '/home/kuru/Desktop/veri-gms-master_noise/spanningtree_veri_pure/'
gpu_id = 0, 1
vis_rank = True
query_remove = True
evaluate = False
dataset_kwargs = {
'source_names': source,
'target_names': target,
'root': root,
'height': height,
'width': width,
'train_batch_size': train_batch_size,
'test_batch_size': test_batch_size,
'train_sampler': train_sampler,
'random_erase': random_erase,
'color_jitter': jitter,
'color_aug': aug
}
transform_kwargs = {
'height': height,
'width': width,
'random_erase': random_erase,
'color_jitter': jitter,
'color_aug': aug
}
optimizer_kwargs = {
'optim': opt,
'lr': lr,
'weight_decay': weight_decay,
'momentum': momentum,
'sgd_dampening': sgd_damp,
'sgd_nesterov': nesterov
}
lr_scheduler_kwargs = {
'lr_scheduler': lr_scheduler,
'stepsize': stepsize,
'gamma': gamma
}
use_gpu = torch.cuda.is_available()
log_name = 'log_test.txt' if evaluate else 'log_train.txt'
sys.stdout = Logger(osp.join(save_dir, log_name))
print('Currently using GPU ', gpu_id)
cudnn.benchmark = True
print('Initializing image data manager')
#dataset = init_imgreid_dataset(root='/home/kuru/Desktop/veri-gms-master/', name='veri')
dataset = init_imgreid_dataset(root='/home/kuru/Desktop/veri-gms-master/',
name='verispan')
train = []
num_train_pids = 0
num_train_cams = 0
print(len(dataset.train))
for img_path, pid, camid, subid, countid in dataset.train:
#print(img_path)
path = img_path[56:90 + 6]
#print(path)
folder = path[1:4]
#print(folder)
#print(img_path, pid, camid,subid,countid)
pid += num_train_pids
camid += num_train_cams
newidd = 0
train.append((path, folder, pid, camid, subid, countid))
#print(train)
#break
num_train_pids += dataset.num_train_pids
num_train_cams += dataset.num_train_cams
pid = 0
pidx = {}
for img_path, pid, camid, subid, countid in dataset.train:
path = img_path[56:90 + 6]
folder = path[1:4]
pidx[folder] = pid
pid += 1
#print(pidx)
sub = []
final = 0
xx = dataset.train
newids = []
print(train[0:2])
train2 = {}
for k in range(0, 770):
for img_path, pid, camid, subid, countid in dataset.train:
if k == pid:
newid = final + subid
sub.append(newid)
#print(pid,subid,newid)
newids.append(newid)
train2[img_path] = newid
#print(img_path, pid, camid, subid, countid, newid)
final = max(sub)
#print(final)
print(len(newids), final)
#train=train2
#print(train2)
train3 = []
for img_path, pid, camid, subid, countid in dataset.train:
#print(img_path,pid,train2[img_path])
path = img_path[56:90 + 6]
#print(path)
folder = path[1:4]
newid = train2[img_path]
#print((path, folder, pid, camid, subid, countid,newid ))
train3.append((path, folder, pid, camid, subid, countid, newid))
train = train3
# for (path, folder, pid, camid, subid, countid,newid) in train:
# print(path, folder)
#path = '/home/kuru/Desktop/adhi/veri-final-draft-master_noise/gmsNoise776/'
path = '/home/kuru/Desktop/veri-gms-master/gms/'
pkl = {}
#pkl[0] = pickle.load('/home/kuru/Desktop/veri-gms-master/gms/620.pkl')
entries = os.listdir(path)
for name in entries:
f = open((path + name), 'rb')
ccc = (path + name)
#print(ccc)
if name == 'featureMatrix.pkl':
s = name[0:13]
else:
s = name[0:3]
#print(s)
#with open (ccc,"rb") as ff:
# pkl[s] = pickle.load(ff)
#print(pkl[s])
pkl[s] = pickle.load(f)
f.close
#print(len(pkl))
print('=> pickle indexing')
data_index = search(pkl)
print(len(data_index))
transform_t = train_transforms(**transform_kwargs)
#print(train[0],train[10])
#data_tfr = vd(pkl_file='index.pkl', dataset = train, root_dir='/home/kuru/Desktop/veri-gms-master/VeRi/image_train/', transform=transform_t)
data_tfr = vdspan(
pkl_file='index_veryspan.pkl',
dataset=train,
root_dir='/home/kuru/Desktop/veri-gms-master/VeRispan/image_train/',
transform=transform_t)
#print(data_tfr)
#print(trainloader)
#data_tfr2=list(data_tfr)
print("lllllllllllllllllllllllllllllllllllllllllllline 433")
df2 = []
data_tfr_old = data_tfr
for (img, label, index, pid, cid, subid, countid, newid) in data_tfr:
#print((img,label,index,pid, cid,subid,countid,newid) )
#print("datframe",(label))
#print(countid)
if countid > 4:
#print(countid)
df2.append((img, label, index, pid, cid, subid, countid, newid))
print("filtered final trainset length", len(df2))
data_tfr = df2
# with open('df2noise_ex.pkl', 'wb') as handle:
# b = pickle.dump(df2, handle, protocol=pickle.HIGHEST_PROTOCOL)
# with open('df2noise.pkl', 'rb') as handle:
# df2 = pickle.load(handle)
# data_tfr=df2
# for (img,label,index,pid, cid,subid,countid,newid) in data_tfr :
# print("datframe",(label))
#data_tfr = vdspansort( dataset = train, root_dir='/home/kuru/Desktop/veri-gms-master_noise/VeRispan/image_train/', transform=transform_t)
#trainloader = DataLoader(df2, sampler=None,batch_size=train_batch_size, shuffle=True, num_workers=workers,pin_memory=True, drop_last=True)
trainloader = DataLoader(data_tfr,
sampler=None,
batch_size=train_batch_size,
shuffle=True,
num_workers=workers,
pin_memory=True,
drop_last=True)
for batch_idx, (img, label, index, pid, cid, subid, countid,
newid) in enumerate(trainloader):
#print("trainloader",batch_idx, (label,index,pid, cid,subid,countid,newid))
print("trainloader", batch_idx, (label))
break
print('Initializing test data manager')
dm = ImageDataManager(use_gpu, **dataset_kwargs)
testloader_dict = dm.return_dataloaders()
print('Initializing model: {}'.format(arch))
model = models.init_model(name=arch,
num_classes=num_train_pids,
loss={'xent', 'htri'},
pretrained=not no_pretrained,
last_stride=2)
print('Model size: {:.3f} M'.format(count_num_param(model)))
if load_weights is not None:
print("weights loaded")
load_pretrained_weights(model, load_weights)
#checkpoint = torch.load('/home/kuru/Desktop/veri-gms-master/logg/model.pth.tar-19')
#model._load_from_state_dict(checkpoint['state_dict'])
#model.load_state_dict(checkpoint['state_dict'])
#optimizer.load_state_dict(checkpoint['optimizer'])
#print(checkpoint['epoch'])
#print(checkpoint['rank1'])
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
print(torch.cuda.device_count())
model = nn.DataParallel(model).cuda() if use_gpu else model
optimizer = init_optimizer(model, **optimizer_kwargs)
#optimizer = init_optimizer(model)
#optimizer.load_state_dict(checkpoint['optimizer'])
scheduler = init_lr_scheduler(optimizer, **lr_scheduler_kwargs)
# scheduler = WarmupMultiStepLR(optimizer, STEPS, GAMMA,
# WARMUP_FACTOR,
# WARMUP_EPOCHS, WARMUP_METHOD)
criterion_xent = CrossEntropyLoss(num_classes=num_train_pids,
use_gpu=use_gpu,
label_smooth=True)
criterion_htri = TripletLoss(margin=margin)
ranking_loss = nn.MarginRankingLoss(margin=margin)
if evaluate:
print('Evaluate only')
for name in target:
print('Evaluating {} ...'.format(name))
queryloader = testloader_dict[name]['query']
galleryloader = testloader_dict[name]['gallery']
_, distmat = test(model,
queryloader,
galleryloader,
train_batch_size,
use_gpu,
return_distmat=True)
if vis_rank:
visualize_ranked_results(distmat,
dm.return_testdataset_by_name(name),
save_dir=osp.join(
save_dir, 'ranked_results', name),
topk=20)
return
time_start = time.time()
ranklogger = RankLogger(source, target)
# # checkpoint = torch.load('/home/kuru/Desktop/market_all/ibna_model/model.pth.tar-79')
# # model.load_state_dict(checkpoint['state_dict'])
# # optimizer.load_state_dict(checkpoint['optimizer'])
# # print(checkpoint['epoch'])
# # start_epoch=checkpoint['epoch']
# # start=start_epoch
# checkpoint = torch.load('/home/kuru/Desktop/veri-gms-master/spanningtreeveri/model.pth.tar-2')
# model.load_state_dict(checkpoint['state_dict'])
# optimizer.load_state_dict(checkpoint['optimizer'])
# print(checkpoint['epoch'])
# start_epoch=checkpoint['epoch']
# start=start_epoch
##start_epoch=resume_from_checkpoint('/home/kuru/Desktop/veri-gms-master/logg/model.pth.tar-20', model, optimizer=None)
print('=> Start training')
for epoch in range(start, max_epoch):
print(epoch, scheduler.get_lr()[0])
#print( torch.cuda.memory_allocated(0))
losses = AverageMeter()
#xent_losses = AverageMeter()
htri_losses = AverageMeter()
accs = AverageMeter()
batch_time = AverageMeter()
xent_losses = AverageMeter()
model.train()
for p in model.parameters():
p.requires_grad = True # open all layers
end = time.time()
for batch_idx, (img, label, index, pid, cid, subid, countid,
newid) in enumerate(trainloader):
trainX, trainY = torch.zeros(
(train_batch_size * 3, 3, height, width),
dtype=torch.float32), torch.zeros((train_batch_size * 3),
dtype=torch.int64)
#pids = torch.zeros((batch_size*3), dtype = torch.int16)
#batchcount=0
for i in range(train_batch_size):
if (countid[i] > 4):
#batchcount=batchcount+1
#print("dfdsfs")
labelx = label[i]
indexx = index[i]
cidx = pid[i]
if indexx > len(pkl[labelx]) - 1:
indexx = len(pkl[labelx]) - 1
#maxx = np.argmax(pkl[labelx][indexx])
a = pkl[labelx][indexx]
minpos = np.argmin(ma.masked_where(a == 0, a))
# print(len(a))
# print(a)
# print(ma.masked_where(a==0, a))
# print(labelx,index,pid,cidx,minpos)
# print(np.array(data_index).shape)
# print(data_index[cidx][1])
pos_dic = data_tfr_old[data_index[cidx][1] + minpos]
#print('posdic', pos_dic)
neg_label = int(labelx)
while True:
neg_label = random.choice(range(1, 770))
#print(neg_label)
if neg_label is not int(labelx) and os.path.isdir(
os.path.join(
'/home/kuru/Desktop/veri-gms-master_noise/veriNoise_train_spanning_folder',
strint(neg_label))) is True:
break
negative_label = strint(neg_label)
neg_cid = pidx[negative_label]
neg_index = random.choice(
range(0, len(pkl[negative_label])))
#print(negative_label,neg_cid,neg_index,data_index[neg_cid] )
neg_dic = data_tfr_old[data_index[neg_cid][1] + neg_index]
#print('negdic', neg_dic)
trainX[i] = img[i]
trainX[i + train_batch_size] = pos_dic[0]
trainX[i + (train_batch_size * 2)] = neg_dic[0]
trainY[i] = cidx
trainY[i + train_batch_size] = pos_dic[3]
trainY[i + (train_batch_size * 2)] = neg_dic[3]
# trainY[i+train_batch_size] = pos_dic[7]
# trainY[i+(train_batch_size*2)] = neg_dic[7]
#break
# else:
# print("skiped",countid[i],subid[i],label[i])
#break
#print(batchcount)
trainX = trainX.cuda()
trainY = trainY.cuda()
outputs, features = model(trainX)
xent_loss = criterion_xent(outputs[0:train_batch_size],
trainY[0:train_batch_size])
htri_loss = criterion_htri(features, trainY)
centerloss = CENTER_LOSS_WEIGHT * center_criterion(
features, trainY)
#tri_loss = ranking_loss(features)
#ent_loss = xent_loss(outputs[0:batch_size], trainY[0:batch_size], num_train_pids)
loss = htri_loss + xent_loss + centerloss
loss = htri_loss + xent_loss
optimizer.zero_grad()
optimizer_center.zero_grad()
loss.backward()
optimizer.step()
# for param in center_criterion.parameters():
# param.grad.data *= (1. /CENTER_LOSS_WEIGHT)
# optimizer_center.step()
for param_group in optimizer.param_groups:
#print(param_group['lr'] )
lrrr = str(param_group['lr'])
batch_time.update(time.time() - end)
losses.update(loss.item(), trainY.size(0))
htri_losses.update(htri_loss.item(), trainY.size(0))
xent_losses.update(xent_loss.item(), trainY.size(0))
accs.update(
accuracy(outputs[0:train_batch_size],
trainY[0:train_batch_size])[0])
if (batch_idx) % 50 == 0:
print('Train ', end=" ")
print('Epoch: [{0}][{1}/{2}]\t'
'Time {batch_time.val:.3f} ({batch_time.avg:.3f})\t'
'TriLoss {loss.val:.4f} ({loss.avg:.4f})\t'
'XLoss {xloss.val:.4f} ({xloss.avg:.4f})\t'
'OveralLoss {oloss.val:.4f} ({oloss.avg:.4f})\t'
'Acc {acc.val:.2f} ({acc.avg:.2f})\t'
'lr {lrrr} \t'.format(
epoch + 1,
batch_idx + 1,
len(trainloader),
batch_time=batch_time,
loss=htri_losses,
xloss=xent_losses,
oloss=losses,
acc=accs,
lrrr=lrrr,
))
end = time.time()
# del loss
# del htri_loss
# del xent_loss
# del htri_losses
# del losses
# del outputs
# del features
# del accs
# del trainX
# del trainY
scheduler.step()
print('=> Test')
save_checkpoint(
{
'state_dict': model.state_dict(),
#'rank1': rank1,
'epoch': epoch + 1,
'arch': arch,
'optimizer': optimizer.state_dict(),
},
save_dir)
GPUtil.showUtilization()
print(torch.cuda.memory_allocated(), torch.cuda.memory_cached())
for name in target:
print('Evaluating {} ...'.format(name))
queryloader = testloader_dict[name]['query']
galleryloader = testloader_dict[name]['gallery']
rank1, distmat = test(model, queryloader, galleryloader,
test_batch_size, use_gpu)
ranklogger.write(name, epoch + 1, rank1)
rank2, distmat2 = test_rerank(model, queryloader, galleryloader,
test_batch_size, use_gpu)
ranklogger.write(name, epoch + 1, rank2)
del queryloader
del galleryloader
del distmat
print(torch.cuda.memory_allocated(), torch.cuda.memory_cached())
torch.cuda.empty_cache()
if (epoch + 1) == max_epoch:
#if (epoch + 1) % 10 == 0:
print('=> Test')
save_checkpoint(
{
'state_dict': model.state_dict(),
'rank1': rank1,
'epoch': epoch + 1,
'arch': arch,
'optimizer': optimizer.state_dict(),
}, save_dir)
for name in target:
print('Evaluating {} ...'.format(name))
queryloader = testloader_dict[name]['query']
galleryloader = testloader_dict[name]['gallery']
rank1, distmat = test(model, queryloader, galleryloader,
test_batch_size, use_gpu)
ranklogger.write(name, epoch + 1, rank1)
# del queryloader
# del galleryloader
# del distmat
if vis_rank:
visualize_ranked_results(
distmat,
dm.return_testdataset_by_name(name),
save_dir=osp.join(save_dir, 'ranked_results', name),
topk=20)
def train_epoch(self):
if self.epoch % self.val_epoch == 0 or self.epoch == 1:
self.validate()
self.model.train()
train_metrics = runningScore(self.n_classes)
train_loss_meter = averageMeter()
self.optim.zero_grad()
for rgb, ir, target in tqdm.tqdm(
self.train_loader, total=len(self.train_loader),
desc=f'Train epoch={self.epoch}', ncols=80, leave=False):
self.iter += 1
assert self.model.training
rgb, ir, target = rgb.to(self.device), ir.to(self.device), target.to(self.device)
score = self.model(rgb, ir)
# score = self.model(rgb)
weight = self.train_loader.dataset.class_weight
if weight:
weight = torch.Tensor(weight).to(self.device)
loss = CrossEntropyLoss(score, target, weight=weight, ignore_index=-1, reduction='mean')
loss_data = loss.data.item()
train_loss_meter.update(loss_data)
if np.isnan(loss_data):
raise ValueError('loss is nan while training')
# loss.backward(retain_graph=True)
loss.backward()
self.optim.step()
self.optim.zero_grad()
if isinstance(score, (tuple, list)):
lbl_pred = score[0].data.max(1)[1].cpu().numpy()
else:
lbl_pred = score.data.max(1)[1].cpu().numpy()
lbl_true = target.data.cpu().numpy()
train_metrics.update(lbl_true, lbl_pred)
acc, acc_cls, mean_iou, fwavacc, _ = train_metrics.get_scores()
metrics = [acc, acc_cls, mean_iou, fwavacc]
with open(osp.join(self.out, 'log.csv'), 'a') as f:
elapsed_time = (
datetime.datetime.now(pytz.timezone('UTC')) -
self.timestamp_start).total_seconds()
log = [self.epoch] + [train_loss_meter.avg] + \
metrics + [''] * 5 + [elapsed_time]
log = map(str, log)
f.write(','.join(log) + '\n')
if self.scheduler:
self.scheduler.step()
if self.epoch % self.val_epoch == 0 or self.epoch == 1:
lr = self.optim.param_groups[0]['lr']
print(f'\nCurrent base learning rate of epoch {self.epoch}: {lr:.7f}')
train_loss_meter.reset()
train_metrics.reset()
def main():
#GENERAL
torch.cuda.empty_cache()
root = "/home/kuru/Desktop/veri-gms-master_noise/"
train_dir = '/home/kuru/Desktop/veri-gms-master_noise/VeRispan/image_train/'
source = {'verispan'}
target = {'verispan'}
workers = 4
height = 280
width = 280
train_size = 32
train_sampler = 'RandomSampler'
#AUGMENTATION
random_erase = True
jitter = True
aug = True
#OPTIMIZATION
opt = 'adam'
lr = 0.0003
weight_decay = 5e-4
momentum = 0.9
sgd_damp = 0.0
nesterov = True
warmup_factor = 0.01
warmup_method = 'linear'
#HYPERPARAMETER
max_epoch = 80
start = 0
train_batch_size = 8
test_batch_size = 100
#SCHEDULER
lr_scheduler = 'multi_step'
stepsize = [30, 60]
gamma = 0.1
#LOSS
margin = 0.3
num_instances = 4
lambda_tri = 1
#MODEL
#arch = 'resnet101'
arch='resnet101_ibn_a'
no_pretrained = False
#TEST SETTINGS
load_weights = '/home/kuru/Desktop/veri-gms-master/IBN-Net_pytorch0.4.1/resnet101_ibn_a.pth'
#load_weights = None
start_eval = 0
eval_freq = -1
#MISC
use_gpu = True
print_freq = 10
seed = 1
resume = ''
save_dir = '/home/kuru/Desktop/veri-gms-master_noise/spanningtree_verinoise_101_stride2/'
gpu_id = 0,1
vis_rank = True
query_remove = True
evaluate = False
dataset_kwargs = {
'source_names': source,
'target_names': target,
'root': root,
'height': height,
'width': width,
'train_batch_size': train_batch_size,
'test_batch_size': test_batch_size,
'train_sampler': train_sampler,
'random_erase': random_erase,
'color_jitter': jitter,
'color_aug': aug
}
transform_kwargs = {
'height': height,
'width': width,
'random_erase': random_erase,
'color_jitter': jitter,
'color_aug': aug
}
optimizer_kwargs = {
'optim': opt,
'lr': lr,
'weight_decay': weight_decay,
'momentum': momentum,
'sgd_dampening': sgd_damp,
'sgd_nesterov': nesterov
}
lr_scheduler_kwargs = {
'lr_scheduler': lr_scheduler,
'stepsize': stepsize,
'gamma': gamma
}
use_gpu = torch.cuda.is_available()
log_name = 'log_test.txt' if evaluate else 'log_train.txt'
sys.stdout = Logger(osp.join(save_dir, log_name))
print('Currently using GPU ', gpu_id)
cudnn.benchmark = True
print('Initializing image data manager')
dataset = init_imgreid_dataset(root='/home/kuru/Desktop/veri-gms-master_noise/', name='verispan')
train = []
num_train_pids = 0
num_train_cams = 0
print(len( dataset.train))
for img_path, pid, camid, subid, countid in dataset.train:
#print(img_path)
path = img_path[56+6:90+6]
#print(path)
folder = path[1:4]
#print(folder)
pid += num_train_pids
newidd=0
train.append((path, folder, pid, camid,subid,countid))
num_train_pids += dataset.num_train_pids
num_train_cams += dataset.num_train_cams
pid = 0
pidx = {}
for img_path, pid, camid, subid, countid in dataset.train:
path = img_path[56+6:90+6]
folder = path[1:4]
pidx[folder] = pid
pid+= 1
sub=[]
final=0
xx=dataset.train
newids=[]
print(train[0:2])
train2={}
for k in range(0,770):
for img_path, pid, camid, subid, countid in dataset.train:
if k==pid:
newid=final+subid
sub.append(newid)
#print(pid,subid,newid)
newids.append(newid)
train2[img_path]= newid
#print(img_path, pid, camid, subid, countid, newid)
final=max(sub)
#print(final)
print(len(newids),final)
#train=train2
#print(train2)
train3=[]
for img_path, pid, camid, subid, countid in dataset.train:
#print(img_path,pid,train2[img_path])
path = img_path[56:90+6]
#print(path)
folder = path[1:4]
newid=train2[img_path]
#print((path, folder, pid, camid, subid, countid,newid ))
train3.append((path, folder, pid, camid, subid, countid,newid ))
train = train3
path = '/home/kuru/Desktop/adhi/veri-final-draft-master_noise/gmsNoise776/'
pkl = {}
#pkl[0] = pickle.load('/home/kuru/Desktop/veri-gms-master/gms/620.pkl')
entries = os.listdir(path)
for name in entries:
f = open((path+name), 'rb')
ccc=(path+name)
#print(ccc)
if name=='featureMatrix.pkl':
s = name[0:13]
else:
s = name[0:3]
#print(s)
#with open (ccc,"rb") as ff:
# pkl[s] = pickle.load(ff)
#print(pkl[s])
pkl[s] = pickle.load(f)
f.close
#print(len(pkl))
with open('cids.pkl', 'rb') as handle:
b = pickle.load(handle)
#print(b)
with open('index.pkl', 'rb') as handle:
c = pickle.load(handle)
transform_t = train_transforms(**transform_kwargs)
data_tfr = vdspan(pkl_file='index_veryspan_noise.pkl', dataset = train, root_dir='/home/kuru/Desktop/veri-gms-master_noise/VeRispan/image_train/', transform=transform_t)
print("lllllllllllllllllllllllllllllllllllllllllllline 433")
df2=[]
data_tfr_old=data_tfr
for (img,label,index,pid, cid,subid,countid,newid) in data_tfr :
#print((img,label,index,pid, cid,subid,countid,newid) )
#print("datframe",(label))
#print(countid)
if countid > 4 :
#print(countid)
df2.append((img,label,index,pid, cid,subid,countid,newid))
print("filtered final trainset length",len(df2))
data_tfr=df2
trainloader = DataLoader(data_tfr, sampler=None,batch_size=train_batch_size, shuffle=True, num_workers=workers,pin_memory=True, drop_last=True)
#data_tfr = vd(pkl_file='index.pkl', dataset = train, root_dir=train_dir,transform=transforms.Compose([Rescale(64),RandomCrop(32),ToTensor()]))
#dataloader = DataLoader(data_tfr, batch_size=batch_size, shuffle=False, num_workers=0)
for batch_idx, (img,label,index,pid, cid,subid,countid,newid) in enumerate(trainloader):
#print("trainloader",batch_idx, (label,index,pid, cid,subid,countid,newid))
print("trainloader",batch_idx, (label))
break
print('Initializing test data manager')
dm = ImageDataManager(use_gpu, **dataset_kwargs)
testloader_dict = dm.return_dataloaders()
print('Initializing model: {}'.format(arch))
model = models.init_model(name=arch, num_classes=num_train_pids, loss={'xent', 'htri'},
pretrained=not no_pretrained, last_stride =2 )
print('Model size: {:.3f} M'.format(count_num_param(model)))
if load_weights is not None:
print("weights loaded")
load_pretrained_weights(model, load_weights)
print(torch.cuda.device_count())
model = nn.DataParallel(model).cuda() if use_gpu else model
optimizer = init_optimizer(model, **optimizer_kwargs)
#optimizer = init_optimizer(model)
scheduler = init_lr_scheduler(optimizer, **lr_scheduler_kwargs)
criterion_xent = CrossEntropyLoss(num_classes=num_train_pids, use_gpu=use_gpu, label_smooth=True)
criterion_htri = TripletLoss(margin=margin)
ranking_loss = nn.MarginRankingLoss(margin = margin)
if evaluate:
print('Evaluate only')
for name in target:
print('Evaluating {} ...'.format(name))
queryloader = testloader_dict[name]['query']
galleryloader = testloader_dict[name]['gallery']
_, distmat = test(model, queryloader, galleryloader, train_batch_size, use_gpu, return_distmat=True)
if vis_rank:
visualize_ranked_results(
distmat, dm.return_testdataset_by_name(name),
save_dir=osp.join(save_dir, 'ranked_results', name),
topk=20
)
return
time_start = time.time()
ranklogger = RankLogger(source, target)
print('=> Start training')
data_index = search(pkl)
print(len(data_index))
for epoch in range(start, max_epoch):
losses = AverageMeter()
#xent_losses = AverageMeter()
htri_losses = AverageMeter()
accs = AverageMeter()
batch_time = AverageMeter()
xent_losses=AverageMeter()
model.train()
for p in model.parameters():
p.requires_grad = True # open all layers
end = time.time()
for batch_idx, (img,label,index,pid, cid,subid,countid,newid) in enumerate(trainloader):
trainX, trainY = torch.zeros((train_batch_size*3,3,height, width), dtype=torch.float32), torch.zeros((train_batch_size*3), dtype = torch.int64)
#pids = torch.zeros((batch_size*3), dtype = torch.int16)
for i in range(train_batch_size):
#print("dfdsfs")
labelx = label[i]
indexx = index[i]
cidx = pid[i]
if indexx >len(pkl[labelx])-1:
indexx = len(pkl[labelx])-1
#maxx = np.argmax(pkl[labelx][indexx])
a = pkl[labelx][indexx]
minpos = np.argmin(ma.masked_where(a==0, a))
#print(minpos)
#print(np.array(data_index).shape)
#print(data_index[cidx][1])
pos_dic = data_tfr_old[data_index[cidx][1]+minpos]
neg_label = int(labelx)
while True:
neg_label = random.choice(range(1, 770))
#print(neg_label)
if neg_label is not int(labelx) and os.path.isdir(os.path.join('/home/kuru/Desktop/adiusb/veri-split/train', strint(neg_label))) is True:
break
negative_label = strint(neg_label)
neg_cid = pidx[negative_label]
neg_index = random.choice(range(0, len(pkl[negative_label])))
neg_dic = data_tfr_old[data_index[neg_cid][1]+neg_index]
trainX[i] = img[i]
trainX[i+train_batch_size] = pos_dic[0]
trainX[i+(train_batch_size*2)] = neg_dic[0]
trainY[i] = cidx
trainY[i+train_batch_size] = pos_dic[3]
trainY[i+(train_batch_size*2)] = neg_dic[3]
trainX = trainX.cuda()
trainY = trainY.cuda()
outputs, features = model(trainX)
xent_loss = criterion_xent(outputs[0:train_batch_size], trainY[0:train_batch_size])
htri_loss = criterion_htri(features, trainY)
#tri_loss = ranking_loss(features)
#ent_loss = xent_loss(outputs[0:batch_size], trainY[0:batch_size], num_train_pids)
loss = htri_loss+xent_loss
optimizer.zero_grad()
loss.backward()
optimizer.step()
batch_time.update(time.time() - end)
losses.update(loss.item(), trainY.size(0))
htri_losses.update(htri_loss.item(), trainY.size(0))
xent_losses.update(xent_loss.item(), trainY.size(0))
accs.update(accuracy(outputs[0:train_batch_size], trainY[0:train_batch_size])[0])
if (batch_idx) % 50 == 0:
print('Train ', end=" ")
print('Epoch: [{0}][{1}/{2}]\t'
'Time {batch_time.val:.3f} ({batch_time.avg:.3f})\t'
'TriLoss {loss.val:.4f} ({loss.avg:.4f})\t'
'XLoss {xloss.val:.4f} ({xloss.avg:.4f})\t'
'OveralLoss {oloss.val:.4f} ({oloss.avg:.4f})\t'
'Acc {acc.val:.2f} ({acc.avg:.2f})\t'
'lr {lrrr} \t'.format(
epoch + 1, batch_idx + 1, len(trainloader),
batch_time=batch_time,
loss = htri_losses,
xloss = xent_losses,
oloss = losses,
acc=accs ,
lrrr=lrrr,
))
end = time.time()
scheduler.step()
print('=> Test')
for name in target:
print('Evaluating {} ...'.format(name))
queryloader = testloader_dict[name]['query']
galleryloader = testloader_dict[name]['gallery']
rank1, distmat = test(model, queryloader, galleryloader, test_batch_size, use_gpu)
ranklogger.write(name, epoch + 1, rank1)
rank2, distmat2 = test_rerank(model, queryloader, galleryloader, test_batch_size, use_gpu)
ranklogger.write(name, epoch + 1, rank2)
#if (epoch + 1) == max_epoch:
if (epoch + 1) % 2 == 0:
print('=> Test')
for name in target:
print('Evaluating {} ...'.format(name))
queryloader = testloader_dict[name]['query']
galleryloader = testloader_dict[name]['gallery']
rank1, distmat = test(model, queryloader, galleryloader, test_batch_size, use_gpu)
ranklogger.write(name, epoch + 1, rank1)
# if vis_rank:
# visualize_ranked_results(
# distmat, dm.return_testdataset_by_name(name),
# save_dir=osp.join(save_dir, 'ranked_results', name),
# topk=20)
save_checkpoint({
'state_dict': model.state_dict(),
'rank1': rank1,
'epoch': epoch + 1,
'arch': arch,
'optimizer': optimizer.state_dict(),
}, save_dir)
def main():
args = parse_args()
update_config(cfg, args)
# cudnn related setting
cudnn.benchmark = cfg.CUDNN.BENCHMARK
torch.backends.cudnn.deterministic = cfg.CUDNN.DETERMINISTIC
torch.backends.cudnn.enabled = cfg.CUDNN.ENABLED
# Set the random seed manually for reproducibility.
np.random.seed(cfg.SEED)
torch.manual_seed(cfg.SEED)
torch.cuda.manual_seed_all(cfg.SEED)
# Loss
criterion = CrossEntropyLoss(cfg.MODEL.NUM_CLASSES).cuda()
# model and optimizer
print(f"Definining network with {cfg.MODEL.LAYERS} layers...")
model = Network(cfg.MODEL.INIT_CHANNELS, cfg.MODEL.NUM_CLASSES, cfg.MODEL.LAYERS, criterion, primitives_2,
drop_path_prob=cfg.TRAIN.DROPPATH_PROB)
model = model.cuda()
# weight params
arch_params = list(map(id, model.arch_parameters()))
weight_params = filter(lambda p: id(p) not in arch_params,
model.parameters())
# Optimizer
optimizer = optim.Adam(
weight_params,
lr=cfg.TRAIN.LR
)
# resume && make log dir and logger
if args.load_path and os.path.exists(args.load_path):
checkpoint_file = os.path.join(args.load_path, 'Model', 'checkpoint_best.pth')
assert os.path.exists(checkpoint_file)
checkpoint = torch.load(checkpoint_file)
# load checkpoint
begin_epoch = checkpoint['epoch']
last_epoch = checkpoint['epoch']
model.load_state_dict(checkpoint['state_dict'])
best_acc1 = checkpoint['best_acc1']
optimizer.load_state_dict(checkpoint['optimizer'])
args.path_helper = checkpoint['path_helper']
logger = create_logger(args.path_helper['log_path'])
logger.info("=> loaded checkpoint '{}'".format(checkpoint_file))
else:
exp_name = args.cfg.split('/')[-1].split('.')[0]
args.path_helper = set_path('logs_search', exp_name)
logger = create_logger(args.path_helper['log_path'])
begin_epoch = cfg.TRAIN.BEGIN_EPOCH
best_acc1 = 0.0
last_epoch = -1
logger.info(args)
logger.info(cfg)
# copy model file
this_dir = os.path.dirname(__file__)
shutil.copy2(
os.path.join(this_dir, 'models', cfg.MODEL.NAME + '.py'),
args.path_helper['ckpt_path'])
# Datasets and dataloaders
# The toy dataset is downloaded with 10 items for each partition. Remove the sample_size parameters to use the full toy dataset
asv_train, asv_dev, asv_eval = asv_toys(sample_size=10)
train_dataset = asv_train #MNIST('mydata', transform=totensor, train=True, download=True)
val_dataset = asv_dev #MNIST('mydata', transform=totensor, train=False, download=True)
train_loader = torch.utils.data.DataLoader(
dataset=train_dataset,
batch_size=cfg.TRAIN.BATCH_SIZE,
num_workers=cfg.DATASET.NUM_WORKERS,
pin_memory=True,
shuffle=True,
drop_last=True,
)
print(f'search.py: Train loader of {len(train_loader)} batches')
print(f'Tot train set: {len(train_dataset)}')
val_loader = torch.utils.data.DataLoader(
dataset=val_dataset,
batch_size=cfg.TRAIN.BATCH_SIZE,
num_workers=cfg.DATASET.NUM_WORKERS,
pin_memory=True,
shuffle=True,
drop_last=True,
)
print(f'search.py: Val loader of {len(val_loader)} batches')
print(f'Tot val set {len(val_dataset)}')
test_dataset = asv_eval #MNIST('mydata', transform=totensor, train=False, download=True)
test_loader = torch.utils.data.DataLoader(
dataset=test_dataset,
batch_size=cfg.TRAIN.BATCH_SIZE,
num_workers=cfg.DATASET.NUM_WORKERS,
pin_memory=True,
shuffle=True,
drop_last=True,
)
# training setting
writer_dict = {
'writer': SummaryWriter(args.path_helper['log_path']),
'train_global_steps': begin_epoch * len(train_loader),
'valid_global_steps': begin_epoch // cfg.VAL_FREQ,
}
# training loop
architect = Architect(model, cfg)
lr_scheduler = torch.optim.lr_scheduler.CosineAnnealingLR(
optimizer, cfg.TRAIN.END_EPOCH, cfg.TRAIN.LR_MIN,
last_epoch=last_epoch
)
for epoch in tqdm(range(begin_epoch, cfg.TRAIN.END_EPOCH), desc='search progress'):
model.train()
genotype = model.genotype()
logger.info('genotype = %s', genotype)
if cfg.TRAIN.DROPPATH_PROB != 0:
model.drop_path_prob = cfg.TRAIN.DROPPATH_PROB * epoch / (cfg.TRAIN.END_EPOCH - 1)
train(cfg, model, optimizer, train_loader, val_loader, criterion, architect, epoch, writer_dict)
if epoch % cfg.VAL_FREQ == 0:
# get threshold and evaluate on validation set
acc = validate_identification(cfg, model, test_loader, criterion)
# remember best acc@1 and save checkpoint
is_best = acc > best_acc1
best_acc1 = max(acc, best_acc1)
# save
logger.info('=> saving checkpoint to {}'.format(args.path_helper['ckpt_path']))
save_checkpoint({
'epoch': epoch + 1,
'state_dict': model.state_dict(),
'best_acc1': best_acc1,
'optimizer': optimizer.state_dict(),
'arch': model.arch_parameters(),
'genotype': genotype,
'path_helper': args.path_helper
}, is_best, args.path_helper['ckpt_path'], 'checkpoint_{}.pth'.format(epoch))
lr_scheduler.step(epoch)
def validate(self):
visualizations = []
val_metrics = runningScore(self.n_classes)
val_loss_meter = averageMeter()
with torch.no_grad():
self.model.eval()
for rgb, ir, target in tqdm.tqdm(
self.val_loader, total=len(self.val_loader),
desc=f'Valid epoch={self.epoch}', ncols=80, leave=False):
rgb, ir, target = rgb.to(self.device), ir.to(self.device), target.to(self.device)
score = self.model(rgb, ir)
# score = self.model(rgb)
weight = self.val_loader.dataset.class_weight
if weight:
weight = torch.Tensor(weight).to(self.device)
loss = CrossEntropyLoss(score, target, weight=weight, reduction='mean', ignore_index=-1)
loss_data = loss.data.item()
if np.isnan(loss_data):
raise ValueError('loss is nan while validating')
val_loss_meter.update(loss_data)
rgbs = rgb.data.cpu()
irs = ir.data.cpu()
if isinstance(score, (tuple, list)):
lbl_pred = score[0].data.max(1)[1].cpu().numpy()
else:
lbl_pred = score.data.max(1)[1].cpu().numpy()
lbl_true = target.data.cpu()
for rgb, ir, lt, lp in zip(rgbs, irs, lbl_true, lbl_pred):
rgb, ir, lt = self.val_loader.dataset.untransform(rgb, ir, lt)
val_metrics.update(lt, lp)
if len(visualizations) < 9:
viz = visualize_segmentation(
lbl_pred=lp, lbl_true=lt, img=rgb, ir=ir,
n_classes=self.n_classes, dataloader=self.train_loader)
visualizations.append(viz)
acc, acc_cls, mean_iou, fwavacc, cls_iu = val_metrics.get_scores()
metrics = [acc, acc_cls, mean_iou, fwavacc]
print(f'\nEpoch: {self.epoch}', f'loss: {val_loss_meter.avg}, mIoU: {mean_iou}')
out = osp.join(self.out, 'visualization_viz')
if not osp.exists(out):
os.makedirs(out)
out_file = osp.join(out, 'epoch{:0>5d}.jpg'.format(self.epoch))
scipy.misc.imsave(out_file, get_tile_image(visualizations))
with open(osp.join(self.out, 'log.csv'), 'a') as f:
elapsed_time = (
datetime.datetime.now(pytz.timezone('UTC')) -
self.timestamp_start).total_seconds()
log = [self.epoch] + [''] * 5 + \
[val_loss_meter.avg] + metrics + [elapsed_time]
log = map(str, log)
f.write(','.join(log) + '\n')
mean_iu = metrics[2]
is_best = mean_iu > self.best_mean_iu
if is_best:
self.best_mean_iu = mean_iu
torch.save({
'epoch': self.epoch,
'arch': self.model.__class__.__name__,
'optim_state_dict': self.optim.state_dict(),
'model_state_dict': self.model.state_dict(),
'best_mean_iu': self.best_mean_iu,
}, osp.join(self.out, 'checkpoint.pth.tar'))
if is_best:
shutil.copy(osp.join(self.out, 'checkpoint.pth.tar'),
osp.join(self.out, 'model_best.pth.tar'))
val_loss_meter.reset()
val_metrics.reset()
class_name = self.val_loader.dataset.class_names
if class_name is not None:
for index, value in enumerate(cls_iu.values()):
offset = 20 - len(class_name[index])
print(class_name[index] + ' ' * offset + f'{value * 100:>.2f}')
else:
print("\nyou don't specify class_names, use number instead")
for key, value in cls_iu.items():
print(key, f'{value * 100:>.2f}')
from datatype import Tensor
def to_one_hot(vector: Tensor) -> Tensor:
"""Create one hot encoding of a vector."""
oh = np.zeros((vector.shape[0], vector.max() + 1))
oh[np.arange(vector.shape[0]), vector] = 1
return oh
train = pd.read_csv("mnist_train.csv", header=None).values[:, 1:]
train_label = pd.read_csv("mnist_train.csv", header=None).values[:, 0]
net = Network(layers=[Linear(784, 128), Sigmoid(), Linear(128, 10),])
loss = CrossEntropyLoss()
optim = SGD(5e-2, 0.001)
x_train, x_val, y_train, y_val = train_test_split(
train.astype(np.float32) / 255,
train_label.astype(np.int32),
test_size=0.2,
random_state=42,
) # to_one_hot
y_train = to_one_hot(y_train)
y_val = to_one_hot(y_val)
batch_size = 100
progress_bar = auto.tqdm(range(200))
accuracies: dict = {"train": [], "val": [], "test": []}
acc_train: list = []
acc_val: list = []
def main():
args = parse_args()
update_config(cfg, args)
# cudnn related setting
cudnn.benchmark = cfg.CUDNN.BENCHMARK
torch.backends.cudnn.deterministic = cfg.CUDNN.DETERMINISTIC
torch.backends.cudnn.enabled = cfg.CUDNN.ENABLED
# Set the random seed manually for reproducibility.
np.random.seed(cfg.SEED)
torch.manual_seed(cfg.SEED)
torch.cuda.manual_seed_all(cfg.SEED)
# model and optimizer
model = eval('resnet.{}(num_classes={})'.format(cfg.MODEL.NAME,
cfg.MODEL.NUM_CLASSES))
model = model.cuda()
optimizer = optim.Adam(
model.net_parameters()
if hasattr(model, 'net_parameters') else model.parameters(),
lr=cfg.TRAIN.LR,
)
# Loss
criterion = CrossEntropyLoss(cfg.MODEL.NUM_CLASSES).cuda()
# resume && make log dir and logger
if args.load_path and os.path.exists(args.load_path):
checkpoint_file = os.path.join(args.load_path, 'Model',
'checkpoint_best.pth')
assert os.path.exists(checkpoint_file)
checkpoint = torch.load(checkpoint_file)
# load checkpoint
begin_epoch = checkpoint['epoch']
last_epoch = checkpoint['epoch']
model.load_state_dict(checkpoint['state_dict'])
best_eer = checkpoint['best_eer']
optimizer.load_state_dict(checkpoint['optimizer'])
args.path_helper = checkpoint['path_helper']
logger = create_logger(args.path_helper['log_path'])
logger.info("=> loaded checkpoint '{}'".format(checkpoint_file))
else:
exp_name = args.cfg.split('/')[-1].split('.')[0]
args.path_helper = set_path('logs', exp_name)
logger = create_logger(args.path_helper['log_path'])
begin_epoch = cfg.TRAIN.BEGIN_EPOCH
best_eer = 1.0
last_epoch = -1
logger.info(args)
logger.info(cfg)
logger.info("Number of parameters: {}".format(count_parameters(model)))
# dataloader
train_dataset = DeepSpeakerDataset(Path(cfg.DATASET.DATA_DIR),
cfg.DATASET.SUB_DIR,
cfg.DATASET.PARTIAL_N_FRAMES)
test_dataset_verification = VoxcelebTestset(Path(cfg.DATASET.DATA_DIR),
cfg.DATASET.PARTIAL_N_FRAMES)
train_loader = torch.utils.data.DataLoader(
dataset=train_dataset,
batch_size=cfg.TRAIN.BATCH_SIZE,
num_workers=cfg.DATASET.NUM_WORKERS,
pin_memory=True,
shuffle=True,
drop_last=True,
)
test_loader_verification = torch.utils.data.DataLoader(
dataset=test_dataset_verification,
batch_size=1,
num_workers=cfg.DATASET.NUM_WORKERS,
pin_memory=True,
shuffle=False,
drop_last=False,
)
# training setting
writer_dict = {
'writer': SummaryWriter(args.path_helper['log_path']),
'train_global_steps': begin_epoch * len(train_loader),
'valid_global_steps': begin_epoch // cfg.VAL_FREQ,
}
# training loop
lr_scheduler = torch.optim.lr_scheduler.CosineAnnealingLR(
optimizer,
cfg.TRAIN.END_EPOCH,
cfg.TRAIN.LR_MIN,
last_epoch=last_epoch)
for epoch in tqdm(range(begin_epoch, cfg.TRAIN.END_EPOCH),
desc='train progress'):
model.train()
train_from_scratch(cfg, model, optimizer, train_loader, criterion,
epoch, writer_dict, lr_scheduler)
if epoch % cfg.VAL_FREQ == 0:
eer = validate_verification(cfg, model, test_loader_verification)
# remember best acc@1 and save checkpoint
is_best = eer < best_eer
best_eer = min(eer, best_eer)
# save
logger.info('=> saving checkpoint to {}'.format(
args.path_helper['ckpt_path']))
save_checkpoint(
{
'epoch': epoch + 1,
'state_dict': model.state_dict(),
'best_eer': best_eer,
'optimizer': optimizer.state_dict(),
'path_helper': args.path_helper
}, is_best, args.path_helper['ckpt_path'],
'checkpoint_{}.pth'.format(epoch))
lr_scheduler.step(epoch)
def main():
# Training settings
parser = argparse.ArgumentParser(description='Cifar10 Example')
parser.add_argument('--batch-size', type=int, default=128, metavar='N', help='input batch size for training (default: 128)')
parser.add_argument('--test-batch-size', type=int, default=1000, metavar='N', help='input batch size for testing (default: 1000)')
parser.add_argument('--epochs', type=int, default=25, metavar='N', help='number of epochs to train (default: 25)')
parser.add_argument('--lr', type=float, default=0.1, metavar='LR', help='learning rate (default: 0.1)')
parser.add_argument('--momentum', type=float, default=0.9, metavar='M', help='SGD momentum (default: 0.9)')
parser.add_argument('--model-path', type=str, default='', metavar='M', help='model param path')
parser.add_argument('--loss-type', type=str, default='CE', metavar='L', help='B or CE or F or ICF_CE or ICF_F or CB_CE or CB_F')
parser.add_argument('--beta', type=float, default=0.999, metavar='B', help='Beta for ClassBalancedLoss')
parser.add_argument('--gamma', type=float, default=2.0, metavar='G', help='Gamma for FocalLoss')
parser.add_argument('--no-cuda', action='store_true', default=False, help='disables CUDA training')
parser.add_argument('--seed', type=int, default=1, metavar='S', help='random seed (default: 1)')
parser.add_argument('--log-interval', type=int, default=10, metavar='N', help='how many batches to wait before logging training status')
parser.add_argument('--balanced-data', action='store_true', default=False, help='For sampling rate. Default is Imbalanced-data.')
parser.add_argument('--save-model', action='store_true', default=False, help='For Saving the current Model')
args = parser.parse_args()
# Add the following code anywhere in your machine learning file
experiment = Experiment(api_key="5Yl3Rxz9S3E0PUKQTBpA0QJPi", project_name="imbalanced-cifar-10", workspace="tancoro")
# ブラウザの実験ページを開く
# experiment.display(clear=True, wait=True, new=0, autoraise=True)
# 実験キー(実験を一意に特定するためのキー)の取得
exp_key = experiment.get_key()
print('KEY: ' + exp_key)
# HyperParamの記録
hyper_params = {
'batch_size': args.batch_size,
'epoch': args.epochs,
'learning_rate': args.lr,
'sgd_momentum' : args.momentum,
'model_path' : args.model_path,
'loss_type' : args.loss_type,
'beta' : args.beta,
'gamma' : args.gamma,
'torch_manual_seed': args.seed,
'balanced_data' : args.balanced_data
}
experiment.log_parameters(hyper_params)
use_cuda = not args.no_cuda and torch.cuda.is_available()
print('use_cuda {}'.format(use_cuda))
torch.manual_seed(args.seed)
device = torch.device("cuda" if use_cuda else "cpu")
kwargs = {'num_workers': 1, 'pin_memory': True} if use_cuda else {}
# train dataset
cifar10_train_dataset = datasets.CIFAR10('./data', train=True, download=True,
transform=transforms.Compose([
transforms.RandomCrop(32, padding=4),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010))
]))
# train sampling rate
sampling_rate = {}
if not args.balanced_data:
sampling_rate = {1:0.05, 4:0.05, 6:0.05}
print(sampling_rate)
# train Sampler
train_sampler = ReductionSampler(cifar10_train_dataset, sampling_rate=sampling_rate)
# train loader
train_loader = torch.utils.data.DataLoader(cifar10_train_dataset,
batch_size=args.batch_size, sampler=train_sampler, **kwargs)
# test dataset
cifar10_test_dataset = datasets.CIFAR10('./data', train=False, transform=transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010))
]))
# test majority loader
test_majority_sampler = ReductionSampler(cifar10_test_dataset, sampling_rate={1:0, 4:0, 6:0})
test_majority_loader = torch.utils.data.DataLoader(cifar10_test_dataset,
batch_size=args.test_batch_size, sampler=test_majority_sampler, **kwargs)
# test minority loader
test_minority_sampler = ReductionSampler(cifar10_test_dataset, sampling_rate={0:0, 2:0, 3:0, 5:0, 7:0, 8:0, 9:0})
test_minority_loader = torch.utils.data.DataLoader(cifar10_test_dataset,
batch_size=args.test_batch_size, sampler=test_minority_sampler, **kwargs)
# test alldata loader
test_alldata_loader = torch.utils.data.DataLoader(cifar10_test_dataset, batch_size=args.test_batch_size, shuffle=True, **kwargs)
model = ResNet18().to(device)
# train loss
train_loss = BasicCrossEntropyLoss()
if args.loss_type == 'CE':
train_loss = CrossEntropyLoss(train_sampler.get_data_count_map(), device)
elif args.loss_type == 'F':
train_loss = FocalLoss(train_sampler.get_data_count_map(), device, gamma=args.gamma)
elif args.loss_type == 'ICF_CE':
train_loss = InverseClassFrequencyCrossEntropyLoss(train_sampler.get_data_count_map(), device)
elif args.loss_type == 'ICF_F':
train_loss = InverseClassFrequencyFocalLoss(train_sampler.get_data_count_map(), device, gamma=args.gamma)
elif args.loss_type == 'CB_CE':
train_loss = ClassBalancedCrossEntropyLoss(train_sampler.get_data_count_map(), device, beta=args.beta)
elif args.loss_type == 'CB_F':
train_loss = ClassBalancedFocalLoss(train_sampler.get_data_count_map(), device, beta=args.beta, gamma=args.gamma)
print('Train Loss Type: {}'.format(type(train_loss)))
# load param
if len(args.model_path) > 0:
model.load_state_dict(torch.load(args.model_path))
optimizer = optim.SGD(model.parameters(), lr=args.lr, momentum=args.momentum, weight_decay=5e-4)
# lr = 0.1 if epoch < 15
# lr = 0.01 if 15 <= epoch < 20
# lr = 0.001 if 20 <= epoch < 25
scheduler = optim.lr_scheduler.MultiStepLR(optimizer, milestones=[15,20], gamma=0.1)
for epoch in range(1, args.epochs + 1):
with experiment.train():
experiment.log_current_epoch(epoch)
train(args, model, device, train_loader, len(train_sampler), optimizer, epoch, experiment, lossfunc=train_loss)
with experiment.test():
test(args, model, device, test_minority_loader, len(test_minority_sampler), epoch, experiment, pref='minority')
test(args, model, device, test_majority_loader, len(test_majority_sampler), epoch, experiment, pref='majority')
test(args, model, device, test_alldata_loader, len(test_alldata_loader.dataset), epoch, experiment, pref='all')
if (args.save_model) and (epoch % 10 == 0):
print('saving model to ./model/cifar10_{0}_{1:04d}.pt'.format(exp_key, epoch))
torch.save(model.state_dict(), "./model/cifar10_{0}_{1:04d}.pt".format(exp_key, epoch))
scheduler.step()
def main():
args = parse_args()
update_config(cfg, args)
# cudnn related setting
cudnn.benchmark = cfg.CUDNN.BENCHMARK
torch.backends.cudnn.deterministic = cfg.CUDNN.DETERMINISTIC
torch.backends.cudnn.enabled = cfg.CUDNN.ENABLED
# Set the random seed manually for reproducibility.
np.random.seed(cfg.SEED)
torch.manual_seed(cfg.SEED)
torch.cuda.manual_seed_all(cfg.SEED)
# Loss
criterion = CrossEntropyLoss(cfg.MODEL.NUM_CLASSES).cuda()
# load arch
genotype = eval(args.text_arch)
model = Network(cfg.MODEL.INIT_CHANNELS, cfg.MODEL.NUM_CLASSES,
cfg.MODEL.LAYERS, genotype)
model = model.cuda()
optimizer = optim.Adam(
model.parameters(),
lr=cfg.TRAIN.LR,
weight_decay=cfg.TRAIN.WD,
)
# resume && make log dir and logger
if args.load_path and os.path.exists(args.load_path):
checkpoint_file = os.path.join(args.load_path, 'Model',
'checkpoint_best.pth')
assert os.path.exists(checkpoint_file)
checkpoint = torch.load(checkpoint_file)
# load checkpoint
begin_epoch = checkpoint['epoch']
last_epoch = checkpoint['epoch']
model.load_state_dict(checkpoint['state_dict'])
best_acc1 = checkpoint['best_acc1']
optimizer.load_state_dict(checkpoint['optimizer'])
args.path_helper = checkpoint['path_helper']
logger = create_logger(args.path_helper['log_path'])
logger.info("=> loaded checkloggpoint '{}'".format(checkpoint_file))
else:
exp_name = args.cfg.split('/')[-1].split('.')[0]
args.path_helper = set_path('logs_scratch', exp_name)
logger = create_logger(args.path_helper['log_path'])
begin_epoch = cfg.TRAIN.BEGIN_EPOCH
best_acc1 = 0.0
last_epoch = -1
logger.info(args)
logger.info(cfg)
logger.info(f"selected architecture: {genotype}")
logger.info("Number of parameters: {}".format(count_parameters(model)))
# copy model file
this_dir = os.path.dirname(__file__)
shutil.copy2(os.path.join(this_dir, './models', cfg.MODEL.NAME + '.py'),
args.path_helper['ckpt_path'])
# dataloader
train_dataset = DeepSpeakerDataset(Path(cfg.DATASET.DATA_DIR),
cfg.DATASET.PARTIAL_N_FRAMES, 'train')
train_loader = torch.utils.data.DataLoader(
dataset=train_dataset,
batch_size=cfg.TRAIN.BATCH_SIZE,
num_workers=cfg.DATASET.NUM_WORKERS,
pin_memory=True,
shuffle=True,
drop_last=True,
)
test_dataset = DeepSpeakerDataset(Path(cfg.DATASET.DATA_DIR),
cfg.DATASET.PARTIAL_N_FRAMES,
'test',
is_test=True)
test_loader = torch.utils.data.DataLoader(
dataset=test_dataset,
batch_size=1,
num_workers=cfg.DATASET.NUM_WORKERS,
pin_memory=True,
shuffle=True,
drop_last=True,
)
# training setting
writer_dict = {
'writer': SummaryWriter(args.path_helper['log_path']),
'train_global_steps': begin_epoch * len(train_loader),
'valid_global_steps': begin_epoch // cfg.VAL_FREQ,
}
# training loop
lr_scheduler = torch.optim.lr_scheduler.CosineAnnealingLR(
optimizer,
cfg.TRAIN.END_EPOCH,
cfg.TRAIN.LR_MIN,
last_epoch=last_epoch)
for epoch in tqdm(range(begin_epoch, cfg.TRAIN.END_EPOCH),
desc='train progress'):
model.train()
model.drop_path_prob = cfg.MODEL.DROP_PATH_PROB * epoch / cfg.TRAIN.END_EPOCH
train_from_scratch(cfg, model, optimizer, train_loader, criterion,
epoch, writer_dict)
if epoch % cfg.VAL_FREQ == 0 or epoch == cfg.TRAIN.END_EPOCH - 1:
acc = validate_identification(cfg, model, test_loader, criterion)
# remember best acc@1 and save checkpoint
is_best = acc > best_acc1
best_acc1 = max(acc, best_acc1)
# save
logger.info('=> saving checkpoint to {}'.format(
args.path_helper['ckpt_path']))
save_checkpoint(
{
'epoch': epoch + 1,
'state_dict': model.state_dict(),
'best_acc1': best_acc1,
'optimizer': optimizer.state_dict(),
'path_helper': args.path_helper,
'genotype': genotype,
}, is_best, args.path_helper['ckpt_path'],
'checkpoint_{}.pth'.format(epoch))
lr_scheduler.step(epoch)
def mnist_model():
model = network.Network()
model.add(FullyConnected(in_feature=784, out_feature=512), name='fc1')
model.add(LeakyReLU(), name='leaky_relu1')
model.add(FullyConnected(in_feature=512, out_feature=256), name='fc2')
model.add(LeakyReLU(), name='leaky_relu2')
model.add(FullyConnected(in_feature=256, out_feature=256), name='fc3')
model.add(LeakyReLU(), name='leaky_relu3')
model.add(FullyConnected(in_feature=256, out_feature=128), name='fc4')
model.add(LeakyReLU(), name='leaky_relu4')
model.add(FullyConnected(in_feature=128, out_feature=10), name='fc5')
model.add(Softmax(), name='softmax')
model.add_loss(CrossEntropyLoss())
optimizer = SGD(lr=1e-4)
print(model)
traingset = fetch_traingset()
train_images, train_labels = traingset['images'], traingset['labels']
batch_size = 256
training_size = len(train_images)
loss_list = np.zeros((50, int(training_size / batch_size)))
for epoch in range(50):
for i in range(int(training_size / batch_size)):
batch_images = np.array(train_images[i * batch_size:(i + 1) *
batch_size])
batch_labels = np.array(train_labels[i * batch_size:(i + 1) *
batch_size])
batch_labels = one_hot(batch_labels, 10)
_, loss = model.forward(batch_images, batch_labels)
if i % 50 == 0:
loss_list[epoch][i] = loss
print("e:{}, i:{} loss: {}".format(epoch, i, loss))
model.backward()
model.optimize(optimizer)
filename = 'model.data'
f = open(filename, 'wb')
pickle.dump(model, f)
f.close()
loss_fname = 'loss.data'
f = open(loss_fname, 'wb')
pickle.dump(loss_list, f)
f.close()
testset = fetch_testingset()
test_images, test_labels = testset['images'], testset['labels']
test_images = np.array(test_images[:])
test_labels_one_hot = one_hot(test_labels, 10)
y_, test_loss = model.forward(test_images, test_labels_one_hot)
test_labels_pred = testResult2labels(y_)
test_labels = np.array(test_labels)
right_num = np.sum(test_labels == test_labels_pred)
accuracy = 1.0 * right_num / test_labels.shape[0]
print('test accuracy is: ', accuracy)
a = 0
def train(network_name='unet', nb_of_epochs=50):
context = None
if network_name == 'unet':
network = UNet(input_dimensions=np.array([56, 56, 56, 1]),
nb_of_classes=2,
depth=2)
batch_manager = DeepMedicBatchManager(input_tile_dimensions=network.input_dimensions,
output_tile_dimensions=network.output_dimensions[:-1],
training_directory='D:\untitled\ku\new',
nb_of_training_images=24,
nb_of_training_images_to_load=18,
nb_of_tiles_per_training_image=1000,
nb_of_tiles_per_training_batch=10,
nb_of_testing_images=6,
nb_of_tiles_per_testing_image=1000,
nb_of_tiles_per_testing_batch=20,
tile_foreground_to_background_ratio=0.2)
trainer = Trainer(initial_learning_rate=0.0005,
loss_function=CrossEntropyLoss())
context = Context(network=network,
batch_manager=batch_manager,
trainer=trainer,
nb_of_training_tiles_before_evaluation=1800)
elif network_name == 'vnet':
network = VNet(input_dimensions=np.array([64, 64, 64, 1]),
nb_of_classes=2,
depth=4)
batch_manager = DeepMedicBatchManager(input_tile_dimensions=network.input_dimensions,
output_tile_dimensions=network.output_dimensions[:-1],
training_directory='D:\untitled\ku\new',
nb_of_training_images=24,
nb_of_training_images_to_load=18,
nb_of_tiles_per_training_image=1000,
nb_of_tiles_per_training_batch=10,
nb_of_testing_images=6,
nb_of_tiles_per_testing_image=1000,
nb_of_tiles_per_testing_batch=20,
tile_foreground_to_background_ratio=0.2)
trainer = Trainer(initial_learning_rate=0.001,
loss_function=CrossEntropyLoss())
context = Context(network=network,
batch_manager=batch_manager,
trainer=trainer,
nb_of_training_tiles_before_evaluation=1800)
elif network_name == 'deepmedic':
network = DeepMedic(input_dimensions=np.array([57, 57, 57, 1]),
sampling_factor=3,
nb_of_classes=2)
batch_manager = DeepMedicBatchManager(input_tile_dimensions=network.input_dimensions,
output_tile_dimensions=network.output_dimensions[:-1],
training_directory='D:\untitled\ku\new',
nb_of_training_images=24,
nb_of_training_images_to_load=18,
nb_of_tiles_per_training_image=1000,
nb_of_tiles_per_training_batch=200, # 20 works well, let's see for 40
nb_of_testing_images=6,
nb_of_tiles_per_testing_image=1000,
nb_of_tiles_per_testing_batch=400,
tile_foreground_to_background_ratio=0.2)
trainer = Trainer(initial_learning_rate=0.0005, # 0.00001 works well, but bn allows for much higher learning_rates, so to be checked...
loss_function=CrossEntropyLoss())
context = Context(network=network,
batch_manager=batch_manager,
trainer=trainer,
nb_of_training_tiles_before_evaluation=1800)
context.build_graph()
context.train(nb_of_epochs)
def Train():
### SEED
if args.seed is not None:
SEED = args.seed
torch.manual_seed(SEED)
torch.cuda.manual_seed(SEED)
np.random.seed(SEED)
torch.backends.cudnn.deterministic = True
##### dataset
mean = [0.4914, 0.4822, 0.4465]
std = [0.2023, 0.1994, 0.2010]
normalize = transforms.Normalize(mean, std)
transform_train = transforms.Compose([
# TODO: uncomment these lines for task 3
Padding(padding=4),
RandomCrop(size=32),
RandomFlip(),
transforms.ToTensor(),
normalize,
])
transform_test = transforms.Compose([
transforms.ToTensor(),
normalize,
])
data = getattr(dataset, args.dataset)(batch_size=args.batch_size,
transform_train=transform_train,
transform_test=transform_test)
#####
# model=alexnet(num_classes=args.num_classes)
# Uncomment the following line for task 5
model = ResNet()
model = nn.DataParallel(model)
cudnn.benchmark = True
model = model.cuda()
if args.calculate_flops:
print_model_parm_flops(model)
return
class_count = [0 for i in range(10)]
for i, (inputs, target) in enumerate(data.train):
for t in target:
class_count[t] += 1
dataset_size = sum(class_count)
print('Class Information: ', class_count)
optimizer = torch.optim.SGD(model.parameters(),
lr=args.lr,
momentum=args.momentum,
weight_decay=args.weight_decay,
nesterov=True)
ce = CrossEntropyLoss()
# Un-comment the following statement for task 2
# ce_train = CrossEntropyLoss()
# Un-comment the following statement for task 4
ce_train = CrossEntropyLoss(cls_count=class_count,
dataset_size=dataset_size)
##### lr multi step schedule
def lr_schedule_multistep(epoch):
if epoch < 5:
factor = (epoch + 1) / 5.0
lr = args.lr * (1 / 3.0 * (1 - factor) + factor)
elif epoch < 80:
lr = args.lr
elif epoch < 90:
lr = args.lr * 0.1
else:
lr = args.lr * 0.001
for param_group in optimizer.param_groups:
param_group['lr'] = lr
return lr
# finetune
if args.resume:
checkpoint = torch.load(args.resume)
model.load_state_dict(checkpoint['state_dict'], strict=True)
start_epoch = checkpoint['epoch']
else:
start_epoch = 0
#####Train
end_epoch = args.epochs
best_acc = 0.0
print(model)
for epoch in range(start_epoch, end_epoch):
#####adjust learning rate every epoch begining
lr = lr_schedule_multistep(epoch)
#####
model.train()
total = 0.0
correct = 0.0
total_loss = 0.0
for i, (inputs, target) in enumerate(data.train):
input, target = inputs.cuda(), target.cuda()
logit = model(input)
loss = ce_train(logit, target, epo=epoch)
total_loss += loss
optimizer.zero_grad()
loss.backward()
optimizer.step()
_, predicted = logit.max(1)
total += target.size(0)
correct += predicted.eq(target).sum().item()
acc_n = logit.max(dim=1)[1].eq(target).sum().item()
log(args, 'Train acc_n %.5f lr %.5f' % (acc_n / input.size(0), lr))
log(args, 'loss = %.5f' % (total_loss / total))
model.eval()
total = 0.0
class_num = torch.zeros(args.num_classes).cuda()
correct = torch.zeros(args.num_classes).cuda()
for i, (inputs, target) in enumerate(data.test):
input, target = inputs.cuda(), target.cuda()
logit = model(input)
# loss = ce(logit, target)
_, predicted = logit.max(1)
total += target.size(0)
target_one_hot = F.one_hot(target, args.num_classes)
predict_one_hot = F.one_hot(predicted, args.num_classes)
class_num = class_num + target_one_hot.sum(dim=0).to(torch.float)
correct = correct + (target_one_hot + predict_one_hot
== 2).sum(dim=0).to(torch.float)
acc = correct.sum() / total
acc_classes = correct / class_num
head_acc = acc_classes[:3].mean()
medium_acc = acc_classes[3:6].mean()
tail_acc = acc_classes[6:10].mean()
if best_acc < acc:
best_acc = acc
torch.save({
'state_dict': model.state_dict(),
'epoch': epoch
}, "model_best.pth")
log(
args, "Test epoch=%d acc=%.5f best_acc=%.5f" %
(epoch, correct.sum() / total, best_acc))
log(args, "Test " + str(correct / class_num))
log(
args, "Test head acc:" + str(head_acc) + " medium acc " +
str(medium_acc) + " tail acc " + str(tail_acc) + '\n')
from data_preparation import load_data
from solve_rnn import solve_rnn
import theano.tensor as T
X_train, y_train, X_test, y_test = load_data()
HIDDEN_DIM = 32
INPUT_DIM = 20
OUTPUT_DIM = 10
model = Network()
model.add(RNN('rnn1', HIDDEN_DIM, INPUT_DIM, 0.1)) # output shape: 4 x HIDDEN_DIM
model.add(Linear('fc', HIDDEN_DIM, OUTPUT_DIM, 0.1)) # output shape: 4 x OUTPUT_DIM
model.add(Softmax('softmax'))
loss = CrossEntropyLoss('xent')
optim = SGDOptimizer(0.01, 0.0001, 0.9)
input_placeholder = T.fmatrix('input')
label_placeholder = T.fmatrix('label')
model.compile(input_placeholder, label_placeholder, loss, optim)
MAX_EPOCH = 6
DISP_FREQ = 1000
TEST_FREQ = 10000
solve_rnn(model, X_train, y_train, X_test, y_test,
MAX_EPOCH, DISP_FREQ, TEST_FREQ)
images = image.T.reshape(batch_size, 1, 28, 28)
images = np.asarray(images)
img_viewer_examples(images, labels.tolist()[0], greyscale= True)
model = Model()
model.add(Dense(784, 90))
model.add(ReLU())
model.add(Dense(90, 45))
model.add(ReLU())
model.add(Dense(45, 10))
model.set_loss(CrossEntropyLoss())
optimizer = Adam(model.parameters(), learning_rate = 0.01)
lr_schedular = StepLR(optimizer, step_size = 1, gamma=0.1)
# weights path
path = "./checkpoints/Linear_MINST_weights.sav"
# model = load_weights(path)
epochs = 6
for epoch in range(epochs):
i = 0
for image, label in dataloader:
if epoch == 5:
model.graph()
image = image/255
def main():
#GENERAL
root = "/home/kuru/Desktop/veri-gms-master/"
train_dir = '/home/kuru/Desktop/veri-gms-master/VeRispan/image_train/'
source = {'veri'}
target = {'veri'}
workers = 2
height = 320
width = 320
train_sampler = 'RandomSampler'
#AUGMENTATION
random_erase = True
jitter = True
aug = True
#OPTIMIZATION
opt = 'adam'
lr = 0.0003
weight_decay = 5e-4
momentum = 0.9
sgd_damp = 0.0
nesterov = True
warmup_factor = 0.01
warmup_method = 'linear'
#HYPERPARAMETER
max_epoch = 80
start = 0
train_batch_size = 16
test_batch_size = 50
#SCHEDULER
lr_scheduler = 'multi_step'
stepsize = [30, 60]
gamma = 0.1
#LOSS
margin = 0.3
num_instances = 6
lambda_tri = 1
#MODEL
arch = 'resnet101_ibn_a'
no_pretrained = False
#TEST SETTINGS
load_weights = '/home/kuru/Desktop/veri-gms-master/IBN-Net_pytorch0.4.1/resnet101_ibn_a.pth'
#load_weights = None
start_eval = 0
eval_freq = -1
#MISC
use_gpu = True
use_amp = True
print_freq = 50
seed = 1
resume = ''
save_dir = '/home/kuru/Desktop/veri-gms-master/logapex/'
gpu_id = 0
vis_rank = True
query_remove = True
evaluate = False
dataset_kwargs = {
'source_names': source,
'target_names': target,
'root': root,
'height': height,
'width': width,
'train_batch_size': train_batch_size,
'test_batch_size': test_batch_size,
'train_sampler': train_sampler,
'random_erase': random_erase,
'color_jitter': jitter,
'color_aug': aug
}
transform_kwargs = {
'height': height,
'width': width,
'random_erase': random_erase,
'color_jitter': jitter,
'color_aug': aug
}
optimizer_kwargs = {
'optim': opt,
'lr': lr,
'weight_decay': weight_decay,
'momentum': momentum,
'sgd_dampening': sgd_damp,
'sgd_nesterov': nesterov
}
lr_scheduler_kwargs = {
'lr_scheduler': lr_scheduler,
'stepsize': stepsize,
'gamma': gamma
}
use_gpu = torch.cuda.is_available()
log_name = 'log_test.txt' if evaluate else 'log_train.txt'
sys.stdout = Logger(osp.join(save_dir, log_name))
print('Currently using GPU ', gpu_id)
cudnn.benchmark = True
print('Initializing image data manager')
dataset = init_imgreid_dataset(root='/home/kuru/Desktop/veri-gms-master/',
name='veri')
train = []
num_train_pids = 0
num_train_cams = 0
for img_path, pid, camid in dataset.train:
path = img_path[52:77]
#print(path)
folder = path[1:4]
pid += num_train_pids
camid += num_train_cams
train.append((path, folder, pid, camid))
num_train_pids += dataset.num_train_pids
num_train_cams += dataset.num_train_cams
pid = 0
pidx = {}
for img_path, pid, camid in dataset.train:
path = img_path[52:77]
folder = path[1:4]
pidx[folder] = pid
pid += 1
path = '/home/kuru/Desktop/veri-gms-master/gms/'
pkl = {}
entries = os.listdir(path)
for name in entries:
f = open((path + name), 'rb')
if name == 'featureMatrix.pkl':
s = name[0:13]
else:
s = name[0:3]
pkl[s] = pickle.load(f)
f.close
transform_t = train_transforms(**transform_kwargs)
data_tfr = vd(
pkl_file='index.pkl',
dataset=train,
root_dir='/home/kuru/Desktop/veri-gms-master/VeRi/image_train/',
transform=transform_t)
trainloader = DataLoader(data_tfr,
sampler=None,
batch_size=train_batch_size,
shuffle=True,
num_workers=workers,
pin_memory=False,
drop_last=True)
#data_tfr = vd(pkl_file='index.pkl', dataset = train, root_dir=train_dir,transform=transforms.Compose([Rescale(64),RandomCrop(32),ToTensor()]))
#dataloader = DataLoader(data_tfr, batch_size=batch_size, shuffle=False, num_workers=0)
print('Initializing test data manager')
dm = ImageDataManager(use_gpu, **dataset_kwargs)
testloader_dict = dm.return_dataloaders()
print('Initializing model: {}'.format(arch))
model = models.init_model(name=arch,
num_classes=num_train_pids,
loss={'xent', 'htri'},
last_stride=1,
pretrained=not no_pretrained,
use_gpu=use_gpu)
print('Model size: {:.3f} M'.format(count_num_param(model)))
if load_weights is not None:
print("weights loaded")
load_pretrained_weights(model, load_weights)
model = (model).cuda() if use_gpu else model
#model = nn.DataParallel(model).cuda() if use_gpu else model
optimizer = init_optimizer(model, **optimizer_kwargs)
#optimizer = init_optimizer(model)
model, optimizer = amp.initialize(model,
optimizer,
opt_level="O2",
keep_batchnorm_fp32=True,
loss_scale="dynamic")
model = nn.DataParallel(model).cuda() if use_gpu else model
scheduler = init_lr_scheduler(optimizer, **lr_scheduler_kwargs)
criterion_xent = CrossEntropyLoss(num_classes=num_train_pids,
use_gpu=use_gpu,
label_smooth=True)
criterion_htri = TripletLoss(margin=margin)
ranking_loss = nn.MarginRankingLoss(margin=margin)
if evaluate:
print('Evaluate only')
for name in target:
print('Evaluating {} ...'.format(name))
queryloader = testloader_dict[name]['query']
galleryloader = testloader_dict[name]['gallery']
_, distmat = test(model,
queryloader,
galleryloader,
train_batch_size,
use_gpu,
return_distmat=True)
if vis_rank:
visualize_ranked_results(distmat,
dm.return_testdataset_by_name(name),
save_dir=osp.join(
save_dir, 'ranked_results', name),
topk=20)
return
time_start = time.time()
ranklogger = RankLogger(source, target)
print('=> Start training')
data_index = search(pkl)
for epoch in range(start, max_epoch):
losses = AverageMeter()
#xent_losses = AverageMeter()
htri_losses = AverageMeter()
accs = AverageMeter()
batch_time = AverageMeter()
model.train()
for p in model.parameters():
p.requires_grad = True # open all layers
end = time.time()
for batch_idx, (img, label, index, pid, cid) in enumerate(trainloader):
trainX, trainY = torch.zeros(
(train_batch_size * 3, 3, height, width),
dtype=torch.float32), torch.zeros((train_batch_size * 3),
dtype=torch.int64)
#pids = torch.zeros((batch_size*3), dtype = torch.int16)
for i in range(train_batch_size):
labelx = str(label[i])
indexx = int(index[i])
cidx = int(pid[i])
if indexx > len(pkl[labelx]) - 1:
indexx = len(pkl[labelx]) - 1
#maxx = np.argmax(pkl[labelx][indexx])
a = pkl[labelx][indexx]
minpos = np.argmax(ma.masked_where(a == 0, a))
pos_dic = data_tfr[data_index[cidx][1] + minpos]
neg_label = int(labelx)
while True:
neg_label = random.choice(range(1, 770))
if neg_label is not int(labelx) and os.path.isdir(
os.path.join(
'/home/kuru/Desktop/adiusb/veri-split/train',
strint(neg_label))) is True:
break
negative_label = strint(neg_label)
neg_cid = pidx[negative_label]
neg_index = random.choice(range(0, len(pkl[negative_label])))
neg_dic = data_tfr[data_index[neg_cid][1] + neg_index]
trainX[i] = img[i]
trainX[i + train_batch_size] = pos_dic[0]
trainX[i + (train_batch_size * 2)] = neg_dic[0]
trainY[i] = cidx
trainY[i + train_batch_size] = pos_dic[3]
trainY[i + (train_batch_size * 2)] = neg_dic[3]
#print("anc",labelx,'posdic', pos_dic[1],pos_dic[2],'negdic', neg_dic[1],neg_dic[2])
trainX = trainX.cuda()
trainY = trainY.cuda()
outputs, features = model(trainX)
xent_loss = criterion_xent(outputs[0:train_batch_size],
trainY[0:train_batch_size])
htri_loss = criterion_htri(features, trainY)
#tri_loss = ranking_loss(features)
#ent_loss = xent_loss(outputs[0:batch_size], trainY[0:batch_size], num_train_pids)
loss = htri_loss + xent_loss
optimizer.zero_grad()
if use_amp:
with amp.scale_loss(loss, optimizer) as scaled_loss:
scaled_loss.backward()
else:
loss.backward()
#loss.backward()
optimizer.step()
batch_time.update(time.time() - end)
losses.update(loss.item(), trainY.size(0))
htri_losses.update(htri_loss.item(), trainY.size(0))
accs.update(
accuracy(outputs[0:train_batch_size],
trainY[0:train_batch_size])[0])
if (batch_idx) % print_freq == 0:
print('Train ', end=" ")
print('Epoch: [{0}][{1}/{2}]\t'
'Time {batch_time.val:.3f} ({batch_time.avg:.3f})\t'
'Loss {loss.val:.4f} ({loss.avg:.4f})\t'
'Acc {acc.val:.2f} ({acc.avg:.2f})\t'.format(
epoch + 1,
batch_idx + 1,
len(trainloader),
batch_time=batch_time,
loss=htri_losses,
acc=accs))
end = time.time()
scheduler.step()
print('=> Test')
for name in target:
print('Evaluating {} ...'.format(name))
queryloader = testloader_dict[name]['query']
galleryloader = testloader_dict[name]['gallery']
rank1, distmat = test(model, queryloader, galleryloader,
test_batch_size, use_gpu)
ranklogger.write(name, epoch + 1, rank1)
rank2, distmat2 = test_rerank(model, queryloader, galleryloader,
test_batch_size, use_gpu)
ranklogger.write(name, epoch + 1, rank2)
del queryloader
del galleryloader
del distmat
#print(torch.cuda.memory_allocated(),torch.cuda.memory_cached())
torch.cuda.empty_cache()
if (epoch + 1) == max_epoch:
#if (epoch + 1) % 10 == 0:
print('=> Test')
save_checkpoint(
{
'state_dict': model.state_dict(),
'rank1': rank1,
'epoch': epoch + 1,
'arch': arch,
'optimizer': optimizer.state_dict(),
}, save_dir)
if vis_rank:
visualize_ranked_results(distmat,
dm.return_testdataset_by_name(name),
save_dir=osp.join(
save_dir, 'ranked_results', name),
topk=20)
perEpooch = len(trainLoader)
maxIter = maxEpoch * perEpooch
print('Dataset statistics')
print("data['classWeights']: ", datas['classWeights'])
print('mean and std: ', datas['mean'], datas['std'])
# define loss function, respectively
weight = torch.from_numpy(datas['classWeights'])
# if(lossFunc == 'ohem'):
# min_kept = int(batch_size // len(gpus) * h * w // 16)
# criteria = ProbOhemCrossEntropy2d(use_weight=True, ignore_label=ignore_label, thresh=0.7, min_kept=min_kept)
# elif(lossFunc == 'label_smoothing'):
# criteria = CrossEntropyLoss2dLabelSmooth(weight=weight, ignore_label=ignore_label)
if (lossFunc == 'CrossEntropy'):
criteria = CrossEntropyLoss(weight=weight, ignore_index=ignore_label)
elif (lossFunc == 'LovaszSoftmax'):
criteria = LovaszSoftmax(ignore_index=ignore_label)
elif (lossFunc == 'focal'):
criteria = FocalLoss2d(weight=weight, ignore_index=ignore_label)
else:
raise NotImplementedError(
'We only support CrossEntropy, LovaszSoftmax and focal as loss function.'
)
if use_cuda:
criteria = criteria.to(device)
if torch.cuda.device_count() > 1:
print("torch.cuda.device_count()=", torch.cuda.device_count())
gpu_nums = torch.cuda.device_count()
model = nn.DataParallel(model).to(device) # multi-card data parallel