mmaction2 指南 (7) 自定义运行配置

mmaction2 指南

mmcv文档
mmaction2文档

7. 自定义Runtime Settings

使用pytorch的优化器

optimizer = dict(type='Adam', lr=0.0003, weight_decay=0.0001)
optimizer = dict(type='Adam', lr=0.001, betas=(0.9, 0.999), eps=1e-08, weight_decay=0, amsgrad=False)

自定义优化器

mmaction2 指南 (5) 自定义新模块

附加配置

使用gradient clip 让训练过程平稳

optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))

使用动量momentum加速收敛
CyclicLrUpdater

CyclicMomentumUpdater

lr_config = dict(
    policy='cyclic',
    target_ratio=(10, 1e-4),
    cyclic_times=1,
    step_ratio_up=0.4,
)
momentum_config = dict(
    policy='cyclic',
    target_ratio=(0.85 / 0.95, 1),
    cyclic_times=1,
    step_ratio_up=0.4,
)

自定义训练策略

参考StepLRHook

mmcv支持的学习率更新策略

配置

lr_config = dict(
    policy='CosineAnnealing',
    warmup='linear',
    warmup_iters=1000,
    warmup_ratio=1.0 / 10,
    min_lr_ratio=1e-5)

自定义工作流 Workflow

每次训练后，使用 EvalHook 做评估，或者用 val workflow

Workflow is a list of (phase, epochs) to specify the running order and epochs. By default it is set to be
Workdlow是个列表(模式，训练轮)，默认是 workflow = [('train', 1)]
训练一轮，验证一轮 [('train', 1), ('val', 1)]

Keyword total_epochs in the config only controls the number of training epochs and will not affect the validation workflow.

自定义 Hooks

注册钩子，等待触发

For hooks with the same priority, they will be triggered in the same order as they are registered.

1. 实现新的hook

下面hook中需要自定义before_run, after_run …需要干什么

from mmcv.runner import HOOKS, Hook

@HOOKS.register_module()
class MyHook(Hook):

    def __init__(self, a, b):
        pass

    def before_run(self, runner):
        pass

    def after_run(self, runner):
        pass

    def before_epoch(self, runner):
        pass

    def after_epoch(self, runner):
        pass

    def before_iter(self, runner):
        pass

    def after_iter(self, runner):
        pass

2. 注册

• mmaction/core/utils/__init__.py 加入 from .my_hook import MyHook

• 用户自定义hook的通过custom_imports 配置

custom_imports = dict(imports=['mmaction.core.utils.my_hook'], allow_failed_imports=False)

3. 修改配置

custom_hooks = [
    dict(type='MyHook', a=a_value, b=b_value, priority='NORMAL'(可设置可不设置))
]

使用MMCV定义的hooks

mmcv_hooks = [
    dict(type='MMCVHook', a=a_value, b=b_value, priority='NORMAL')
]

下面是已经有的 mmcv 自带的 hook

• log_config
• checkpoint_config
• evaluation
• lr_config
• optimizer_config
• momentum_config

logger hook 优先级最低 VERY_LOW，其余都是NORMAL

修改 checkpoint hook

参考 CheckpointHook

checkpoint_config = dict(interval=1)

修改 log hook

参考 (mmcv.runner.LoggerHook)[https://mmcv.readthedocs.io/en/latest/api.html#mmcv.runner.LoggerHook]

log_config = dict(
    interval=50,
    hooks=[
        dict(type='TextLoggerHook'),
        dict(type='TensorboardLoggerHook')
    ])

修改 evaluation hook

参考 eval_hooks

evaluation = dict(interval=1, metrics='bbox')