File size: 5,176 Bytes
f549064
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
# Copyright (c) OpenMMLab. All rights reserved.
from mmengine.optim.scheduler.lr_scheduler import LRSchedulerMixin
from mmengine.optim.scheduler.momentum_scheduler import MomentumSchedulerMixin
from mmengine.optim.scheduler.param_scheduler import INF, _ParamScheduler
from torch.optim import Optimizer

from mmdet.registry import PARAM_SCHEDULERS


@PARAM_SCHEDULERS.register_module()
class QuadraticWarmupParamScheduler(_ParamScheduler):
    r"""Warm up the parameter value of each parameter group by quadratic
    formula:

    .. math::

        X_{t} = X_{t-1} + \frac{2t+1}{{(end-begin)}^{2}} \times X_{base}

    Args:
        optimizer (Optimizer): Wrapped optimizer.
        param_name (str): Name of the parameter to be adjusted, such as
            ``lr``, ``momentum``.
        begin (int): Step at which to start updating the parameters.
            Defaults to 0.
        end (int): Step at which to stop updating the parameters.
            Defaults to INF.
        last_step (int): The index of last step. Used for resume without
            state dict. Defaults to -1.
        by_epoch (bool): Whether the scheduled parameters are updated by
            epochs. Defaults to True.
        verbose (bool): Whether to print the value for each update.
            Defaults to False.
    """

    def __init__(self,
                 optimizer: Optimizer,
                 param_name: str,
                 begin: int = 0,
                 end: int = INF,
                 last_step: int = -1,
                 by_epoch: bool = True,
                 verbose: bool = False):
        if end >= INF:
            raise ValueError('``end`` must be less than infinity,'
                             'Please set ``end`` parameter of '
                             '``QuadraticWarmupScheduler`` as the '
                             'number of warmup end.')
        self.total_iters = end - begin
        super().__init__(
            optimizer=optimizer,
            param_name=param_name,
            begin=begin,
            end=end,
            last_step=last_step,
            by_epoch=by_epoch,
            verbose=verbose)

    @classmethod
    def build_iter_from_epoch(cls,
                              *args,
                              begin=0,
                              end=INF,
                              by_epoch=True,
                              epoch_length=None,
                              **kwargs):
        """Build an iter-based instance of this scheduler from an epoch-based
        config."""
        assert by_epoch, 'Only epoch-based kwargs whose `by_epoch=True` can ' \
                         'be converted to iter-based.'
        assert epoch_length is not None and epoch_length > 0, \
            f'`epoch_length` must be a positive integer, ' \
            f'but got {epoch_length}.'
        by_epoch = False
        begin = begin * epoch_length
        if end != INF:
            end = end * epoch_length
        return cls(*args, begin=begin, end=end, by_epoch=by_epoch, **kwargs)

    def _get_value(self):
        """Compute value using chainable form of the scheduler."""
        if self.last_step == 0:
            return [
                base_value * (2 * self.last_step + 1) / self.total_iters**2
                for base_value in self.base_values
            ]

        return [
            group[self.param_name] + base_value *
            (2 * self.last_step + 1) / self.total_iters**2
            for base_value, group in zip(self.base_values,
                                         self.optimizer.param_groups)
        ]


@PARAM_SCHEDULERS.register_module()
class QuadraticWarmupLR(LRSchedulerMixin, QuadraticWarmupParamScheduler):
    """Warm up the learning rate of each parameter group by quadratic formula.

    Args:
        optimizer (Optimizer): Wrapped optimizer.
        begin (int): Step at which to start updating the parameters.
            Defaults to 0.
        end (int): Step at which to stop updating the parameters.
            Defaults to INF.
        last_step (int): The index of last step. Used for resume without
            state dict. Defaults to -1.
        by_epoch (bool): Whether the scheduled parameters are updated by
            epochs. Defaults to True.
        verbose (bool): Whether to print the value for each update.
            Defaults to False.
    """


@PARAM_SCHEDULERS.register_module()
class QuadraticWarmupMomentum(MomentumSchedulerMixin,
                              QuadraticWarmupParamScheduler):
    """Warm up the momentum value of each parameter group by quadratic formula.

    Args:
        optimizer (Optimizer): Wrapped optimizer.
        begin (int): Step at which to start updating the parameters.
            Defaults to 0.
        end (int): Step at which to stop updating the parameters.
            Defaults to INF.
        last_step (int): The index of last step. Used for resume without
            state dict. Defaults to -1.
        by_epoch (bool): Whether the scheduled parameters are updated by
            epochs. Defaults to True.
        verbose (bool): Whether to print the value for each update.
            Defaults to False.
    """