Python gluoncv.utils.LRScheduler() Examples
The following are 10
code examples of gluoncv.utils.LRScheduler().
You can vote up the ones you like or vote down the ones you don't like,
and go to the original project or source file by following the links above each example.
You may also want to check out all available functions/classes of the module
gluoncv.utils
, or try the search function
.
.
Example #1
| Source File: test_lr_scheduler.py From gluon-cv with Apache License 2.0 | 6 votes |
|
def test_sanity():
N = 1000
constant = LRScheduler('constant', base_lr=0, target_lr=1, niters=N)
linear = LRScheduler('linear', base_lr=1, target_lr=2, niters=N)
cosine = LRScheduler('cosine', base_lr=3, target_lr=1, niters=N)
poly = LRScheduler('poly', base_lr=1, target_lr=0, niters=N, power=2)
step = LRScheduler('step', base_lr=1, target_lr=0, niters=N,
step_iter=[100, 500], step_factor=0.1)
compare(constant, 0, 0)
compare(constant, N-1, 0)
compare(linear, 0, 1)
compare(linear, N-1, 2)
compare(cosine, 0, 3)
compare(cosine, N-1, 1)
compare(poly, 0, 1)
compare(poly, N-1, 0)
compare(step, 0, 1)
compare(step, 100, 0.1)
compare(step, 500, 0.01)
compare(step, N-1, 0.01)
Example #2
| Source File: test_lr_scheduler.py From gluon-cv with Apache License 2.0 | 6 votes |
|
def test_params():
N = 1000
linear = LRScheduler('linear', base_lr=1, target_lr=2, niters=N)
linear2 = LRScheduler('linear', baselr=1, targetlr=2, niters=N)
linear3 = LRScheduler('linear', base_lr=1, target_lr=2, niters=N,
baselr=0, targetlr=1)
linear4 = LRScheduler('linear', base_lr=1, target_lr=2, niters=N/2)
linear5 = LRScheduler('linear', base_lr=1, target_lr=2, niters=N/2,
nepochs=N/2, iters_per_epoch=2)
compare(linear, 0, 1)
compare(linear, N-1, 2)
compare(linear2, 0, 1)
compare(linear2, N-1, 2)
compare(linear3, 0, 1)
compare(linear3, N-1, 2)
compare(linear4, 0, 1)
compare(linear4, N/2-1, 2)
compare(linear5, 0, 1)
compare(linear5, N/2-1, 1.5, rtol=0.01)
compare(linear5, N-1, 2)
Example #3
| Source File: test_lr_scheduler.py From panoptic-fpn-gluon with Apache License 2.0 | 6 votes |
|
def test_sanity():
N = 1000
constant = LRScheduler('constant', base_lr=0, target_lr=1, niters=N)
linear = LRScheduler('linear', base_lr=1, target_lr=2, niters=N)
cosine = LRScheduler('cosine', base_lr=3, target_lr=1, niters=N)
poly = LRScheduler('poly', base_lr=1, target_lr=0, niters=N, power=2)
step = LRScheduler('step', base_lr=1, target_lr=0, niters=N,
step_iter=[100, 500], step_factor=0.1)
compare(constant, 0, 0)
compare(constant, N-1, 0)
compare(linear, 0, 1)
compare(linear, N-1, 2)
compare(cosine, 0, 3)
compare(cosine, N-1, 1)
compare(poly, 0, 1)
compare(poly, N-1, 0)
compare(step, 0, 1)
compare(step, 100, 0.1)
compare(step, 500, 0.01)
compare(step, N-1, 0.01)
Example #4
| Source File: test_lr_scheduler.py From panoptic-fpn-gluon with Apache License 2.0 | 6 votes |
|
def test_params():
N = 1000
linear = LRScheduler('linear', base_lr=1, target_lr=2, niters=N)
linear2 = LRScheduler('linear', baselr=1, targetlr=2, niters=N)
linear3 = LRScheduler('linear', base_lr=1, target_lr=2, niters=N,
baselr=0, targetlr=1)
linear4 = LRScheduler('linear', base_lr=1, target_lr=2, niters=N/2)
linear5 = LRScheduler('linear', base_lr=1, target_lr=2, niters=N/2,
nepochs=N/2, iters_per_epoch=2)
compare(linear, 0, 1)
compare(linear, N-1, 2)
compare(linear2, 0, 1)
compare(linear2, N-1, 2)
compare(linear3, 0, 1)
compare(linear3, N-1, 2)
compare(linear4, 0, 1)
compare(linear4, N/2-1, 2)
compare(linear5, 0, 1)
compare(linear5, N/2-1, 1.5, rtol=0.01)
compare(linear5, N-1, 2)
Example #5
| Source File: test_lr_scheduler.py From gluon-cv with Apache License 2.0 | 5 votes |
|
def test_composed_method():
N = 1000
constant = LRScheduler('constant', base_lr=0, target_lr=1, niters=N)
linear = LRScheduler('linear', base_lr=1, target_lr=2, niters=N)
cosine = LRScheduler('cosine', base_lr=3, target_lr=1, niters=N)
poly = LRScheduler('poly', base_lr=1, target_lr=0, niters=N, power=2)
# components with niters=0 will be ignored
null_cosine = LRScheduler('cosine', base_lr=3, target_lr=1, niters=0)
null_poly = LRScheduler('cosine', base_lr=3, target_lr=1, niters=0)
step = LRScheduler('step', base_lr=1, target_lr=0, niters=N,
step_iter=[100, 500], step_factor=0.1)
arr = LRSequential([constant, null_cosine, linear, cosine, null_poly, poly, step])
# constant
for i in range(N):
compare(arr, i, 0)
# linear
for i in range(N, 2*N):
expect_linear = 2 + (1 - 2) * (1 - (i - N) / (N - 1))
compare(arr, i, expect_linear)
# cosine
for i in range(2*N, 3*N):
expect_cosine = 1 + (3 - 1) * ((1 + cos(pi * (i - 2*N) / (N - 1))) / 2)
compare(arr, i, expect_cosine)
# poly
for i in range(3*N, 4*N):
expect_poly = 0 + (1 - 0) * (pow(1 - (i - 3*N) / (N - 1), 2))
compare(arr, i, expect_poly)
for i in range(4*N, 5*N):
if i - 4*N < 100:
expect_step = 1
elif i - 4*N < 500:
expect_step = 0.1
else:
expect_step = 0.01
compare(arr, i, expect_step)
# out-of-bound index
compare(arr, 10*N, 0.01)
compare(arr, -1, 0)
Example #6
| Source File: test_lr_scheduler.py From panoptic-fpn-gluon with Apache License 2.0 | 5 votes |
|
def test_composed_method():
N = 1000
constant = LRScheduler('constant', base_lr=0, target_lr=1, niters=N)
linear = LRScheduler('linear', base_lr=1, target_lr=2, niters=N)
cosine = LRScheduler('cosine', base_lr=3, target_lr=1, niters=N)
poly = LRScheduler('poly', base_lr=1, target_lr=0, niters=N, power=2)
# components with niters=0 will be ignored
null_cosine = LRScheduler('cosine', base_lr=3, target_lr=1, niters=0)
null_poly = LRScheduler('cosine', base_lr=3, target_lr=1, niters=0)
step = LRScheduler('step', base_lr=1, target_lr=0, niters=N,
step_iter=[100, 500], step_factor=0.1)
arr = LRSequential([constant, null_cosine, linear, cosine, null_poly, poly, step])
# constant
for i in range(N):
compare(arr, i, 0)
# linear
for i in range(N, 2*N):
expect_linear = 2 + (1 - 2) * (1 - (i - N) / (N - 1))
compare(arr, i, expect_linear)
# cosine
for i in range(2*N, 3*N):
expect_cosine = 1 + (3 - 1) * ((1 + cos(pi * (i - 2*N) / (N - 1))) / 2)
compare(arr, i, expect_cosine)
# poly
for i in range(3*N, 4*N):
expect_poly = 0 + (1 - 0) * (pow(1 - (i - 3*N) / (N - 1), 2))
compare(arr, i, expect_poly)
for i in range(4*N, 5*N):
if i - 4*N < 100:
expect_step = 1
elif i - 4*N < 500:
expect_step = 0.1
else:
expect_step = 0.01
compare(arr, i, expect_step)
# out-of-bound index
compare(arr, 10*N, 0.01)
compare(arr, -1, 0)
Example #7
| Source File: test_lr_scheduler.py From gluon-cv with Apache License 2.0 | 4 votes |
|
def test_single_method():
N = 1000
constant = LRScheduler('constant', base_lr=0, target_lr=1, niters=N)
linear = LRScheduler('linear', base_lr=1, target_lr=2, niters=N)
cosine = LRScheduler('cosine', base_lr=3, target_lr=1, niters=N)
poly = LRScheduler('poly', base_lr=1, target_lr=0, niters=N, power=2)
step = LRScheduler('step', base_lr=1, target_lr=0, niters=N,
step_iter=[100, 500], step_factor=0.1)
step2 = LRScheduler('step', base_lr=1, target_lr=0,
nepochs=2, iters_per_epoch=N/2,
step_iter=[100, 500], step_factor=0.1)
step3 = LRScheduler('step', base_lr=1, target_lr=0,
nepochs=100, iters_per_epoch=N/100,
step_epoch=[10, 50], step_factor=0.1)
# Test numerical value
for i in range(N):
compare(constant, i, 0)
expect_linear = 2 + (1 - 2) * (1 - i / (N - 1))
compare(linear, i, expect_linear)
expect_cosine = 1 + (3 - 1) * ((1 + cos(pi * i / (N-1))) / 2)
compare(cosine, i, expect_cosine)
expect_poly = 0 + (1 - 0) * (pow(1 - i / (N-1), 2))
compare(poly, i, expect_poly)
if i < 100:
expect_step = 1
elif i < 500:
expect_step = 0.1
else:
expect_step = 0.01
compare(step, i, expect_step)
compare(step2, i, expect_step)
compare(step3, i, expect_step)
# Test out-of-range updates
for i in range(10):
constant.update(i - 3)
linear.update(i - 3)
cosine.update(i - 3)
poly.update(i - 3)
Example #8
| Source File: test_lr_scheduler.py From panoptic-fpn-gluon with Apache License 2.0 | 4 votes |
|
def test_single_method():
N = 1000
constant = LRScheduler('constant', base_lr=0, target_lr=1, niters=N)
linear = LRScheduler('linear', base_lr=1, target_lr=2, niters=N)
cosine = LRScheduler('cosine', base_lr=3, target_lr=1, niters=N)
poly = LRScheduler('poly', base_lr=1, target_lr=0, niters=N, power=2)
step = LRScheduler('step', base_lr=1, target_lr=0, niters=N,
step_iter=[100, 500], step_factor=0.1)
step2 = LRScheduler('step', base_lr=1, target_lr=0,
nepochs=2, iters_per_epoch=N/2,
step_iter=[100, 500], step_factor=0.1)
step3 = LRScheduler('step', base_lr=1, target_lr=0,
nepochs=100, iters_per_epoch=N/100,
step_epoch=[10, 50], step_factor=0.1)
# Test numerical value
for i in range(N):
compare(constant, i, 0)
expect_linear = 2 + (1 - 2) * (1 - i / (N - 1))
compare(linear, i, expect_linear)
expect_cosine = 1 + (3 - 1) * ((1 + cos(pi * i / (N-1))) / 2)
compare(cosine, i, expect_cosine)
expect_poly = 0 + (1 - 0) * (pow(1 - i / (N-1), 2))
compare(poly, i, expect_poly)
if i < 100:
expect_step = 1
elif i < 500:
expect_step = 0.1
else:
expect_step = 0.01
compare(step, i, expect_step)
compare(step2, i, expect_step)
compare(step3, i, expect_step)
# Test out-of-range updates
for i in range(10):
constant.update(i - 3)
linear.update(i - 3)
cosine.update(i - 3)
poly.update(i - 3)
Example #9
| Source File: train.py From panoptic-fpn-gluon with Apache License 2.0 | 4 votes |
|
def __init__(self, args):
self.args = args
# image transform
input_transform = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize([.485, .456, .406], [.229, .224, .225]),
])
# dataset and dataloader
data_kwargs = {'transform': input_transform, 'base_size': args.base_size,
'crop_size': args.crop_size}
trainset = get_segmentation_dataset(
args.dataset, split=args.train_split, mode='train', **data_kwargs)
valset = get_segmentation_dataset(
args.dataset, split='val', mode='val', **data_kwargs)
self.train_data = gluon.data.DataLoader(
trainset, args.batch_size, shuffle=True, last_batch='rollover',
num_workers=args.workers)
self.eval_data = gluon.data.DataLoader(valset, args.test_batch_size,
last_batch='rollover', num_workers=args.workers)
# create network
if args.model_zoo is not None:
model = get_model(args.model_zoo, pretrained=True)
else:
model = get_segmentation_model(model=args.model, dataset=args.dataset,
backbone=args.backbone, norm_layer=args.norm_layer,
norm_kwargs=args.norm_kwargs, aux=args.aux,
crop_size=args.crop_size)
model.cast(args.dtype)
print(model)
self.net = DataParallelModel(model, args.ctx, args.syncbn)
self.evaluator = DataParallelModel(SegEvalModel(model), args.ctx)
# resume checkpoint if needed
if args.resume is not None:
if os.path.isfile(args.resume):
model.load_parameters(args.resume, ctx=args.ctx)
else:
raise RuntimeError("=> no checkpoint found at '{}'" \
.format(args.resume))
# create criterion
criterion = MixSoftmaxCrossEntropyLoss(args.aux, aux_weight=args.aux_weight)
self.criterion = DataParallelCriterion(criterion, args.ctx, args.syncbn)
# optimizer and lr scheduling
self.lr_scheduler = LRScheduler(mode='poly', base_lr=args.lr,
nepochs=args.epochs,
iters_per_epoch=len(self.train_data),
power=0.9)
kv = mx.kv.create(args.kvstore)
optimizer_params = {'lr_scheduler': self.lr_scheduler,
'wd':args.weight_decay,
'momentum': args.momentum}
if args.dtype == 'float16':
optimizer_params['multi_precision'] = True
if args.no_wd:
for k, v in self.net.module.collect_params('.*beta|.*gamma|.*bias').items():
v.wd_mult = 0.0
self.optimizer = gluon.Trainer(self.net.module.collect_params(), 'sgd',
optimizer_params, kvstore = kv)
# evaluation metrics
self.metric = gluoncv.utils.metrics.SegmentationMetric(trainset.num_class)
Example #10
| Source File: train.py From cascade_rcnn_gluon with Apache License 2.0 | 4 votes |
|
def __init__(self, args):
self.args = args
# image transform
input_transform = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize([.485, .456, .406], [.229, .224, .225]),
])
# dataset and dataloader
trainset = get_segmentation_dataset(
args.dataset, split='train', transform=input_transform)
valset = get_segmentation_dataset(
args.dataset, split='val', transform=input_transform)
self.train_data = gluon.data.DataLoader(
trainset, args.batch_size, shuffle=True, last_batch='rollover',
num_workers=args.workers)
self.eval_data = gluon.data.DataLoader(valset, args.test_batch_size,
last_batch='keep', num_workers=args.workers)
# create network
model = get_segmentation_model(model=args.model, dataset=args.dataset,
backbone=args.backbone, norm_layer=args.norm_layer,
aux=args.aux)
print(model)
self.net = DataParallelModel(model, args.ctx, args.syncbn)
self.evaluator = DataParallelModel(SegEvalModel(model), args.ctx)
# resume checkpoint if needed
if args.resume is not None:
if os.path.isfile(args.resume):
model.load_params(args.resume, ctx=args.ctx)
else:
raise RuntimeError("=> no checkpoint found at '{}'" \
.format(args.resume))
# create criterion
criterion = SoftmaxCrossEntropyLossWithAux(args.aux)
self.criterion = DataParallelCriterion(criterion, args.ctx, args.syncbn)
# optimizer and lr scheduling
self.lr_scheduler = LRScheduler(mode='poly', baselr=args.lr, niters=len(self.train_data),
nepochs=args.epochs)
kv = mx.kv.create(args.kvstore)
self.optimizer = gluon.Trainer(self.net.module.collect_params(), 'sgd',
{'lr_scheduler': self.lr_scheduler,
'wd':args.weight_decay,
'momentum': args.momentum,
'multi_precision': True},
kvstore = kv)
