Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

train: allow passing custom learning rate optimizer #305

Open
wants to merge 1 commit into
base: master
Choose a base branch
from
Open

train: allow passing custom learning rate optimizer #305

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
14 changes: 10 additions & 4 deletions train.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -169,7 +169,11 @@ def gather(self, outputs, output_device):

return out

def train():
def train(optimizer=None):
"""
@param optimizer: set custom optimizer, default (None) uses
`torch.optim.SGD(net.parameters(), lr=args.lr, momentum=args.momentum, weight_decay=args.decay)`
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

ideally, I'd set the optimizer=SGD(...) here already, but the "net" is not available

"""
if not os.path.exists(args.save_folder):
os.mkdir(args.save_folder)

Expand Down Expand Up @@ -212,8 +216,10 @@ def train():
print('Initializing weights...')
yolact_net.init_weights(backbone_path=args.save_folder + cfg.backbone.path)

optimizer = optim.SGD(net.parameters(), lr=args.lr, momentum=args.momentum,
if optimizer is None:
optimizer = optim.SGD(net.parameters(), lr=args.lr, momentum=args.momentum,
weight_decay=args.decay)

criterion = MultiBoxLoss(num_classes=cfg.num_classes,
pos_threshold=cfg.positive_iou_threshold,
neg_threshold=cfg.negative_iou_threshold,
Expand Down Expand Up @@ -291,11 +297,11 @@ def train():
if changed:
cfg.delayed_settings = [x for x in cfg.delayed_settings if x[0] > iteration]

# Warm up by linearly interpolating the learning rate from some smaller value
# Warm up by linearly interpolating the learning rate from some smaller value
if cfg.lr_warmup_until > 0 and iteration <= cfg.lr_warmup_until:
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

can we leave this fine tuning, lr management to the optimizer (or other code provided by the framework)?

set_lr(optimizer, (args.lr - cfg.lr_warmup_init) * (iteration / cfg.lr_warmup_until) + cfg.lr_warmup_init)

# Adjust the learning rate at the given iterations, but also if we resume from past that iteration
# Adjust the learning rate at the given iterations, but also if we resume from past that iteration
while step_index < len(cfg.lr_steps) and iteration >= cfg.lr_steps[step_index]:
step_index += 1
set_lr(optimizer, args.lr * (args.gamma ** step_index))
Expand Down