transfiner / configs /common /coco_schedule.py
lkeab
update configs
c6c496f
raw
history blame
1.66 kB
from fvcore.common.param_scheduler import MultiStepParamScheduler
from detectron2.config import LazyCall as L
from detectron2.solver import WarmupParamScheduler
def default_X_scheduler(num_X):
"""
Returns the config for a default multi-step LR scheduler such as "1x", "3x",
commonly referred to in papers, where every 1x has the total length of 1440k
training images (~12 COCO epochs). LR is decayed twice at the end of training
following the strategy defined in "Rethinking ImageNet Pretraining", Sec 4.
Args:
num_X: a positive real number
Returns:
DictConfig: configs that define the multiplier for LR during training
"""
# total number of iterations assuming 16 batch size, using 1440000/16=90000
total_steps_16bs = num_X * 90000
if num_X <= 2:
scheduler = L(MultiStepParamScheduler)(
values=[1.0, 0.1, 0.01],
# note that scheduler is scale-invariant. This is equivalent to
# milestones=[6, 8, 9]
milestones=[60000, 80000, 90000],
)
else:
scheduler = L(MultiStepParamScheduler)(
values=[1.0, 0.1, 0.01],
milestones=[total_steps_16bs - 60000, total_steps_16bs - 20000, total_steps_16bs],
)
return L(WarmupParamScheduler)(
scheduler=scheduler,
warmup_length=1000 / total_steps_16bs,
warmup_method="linear",
warmup_factor=0.001,
)
lr_multiplier_1x = default_X_scheduler(1)
lr_multiplier_2x = default_X_scheduler(2)
lr_multiplier_3x = default_X_scheduler(3)
lr_multiplier_6x = default_X_scheduler(6)
lr_multiplier_9x = default_X_scheduler(9)