# Copyright (c) OpenMMLab. All rights reserved. from typing import Optional, Union from mmengine.hooks import Hook from mmengine.model import is_model_wrapper from mmengine.registry import MODELS from torch import nn from mmdet.registry import HOOKS @HOOKS.register_module() class DataPreprocessorSwitchHook(Hook): """Switch the data_preprocessor during training. Args: switch_epoch (int): switch pipeline at this epoch. switch_data_preprocessor (list[dict]): the data_preprocessor to switch to. """ def __init__( self, switch_epoch, switch_data_preprocessor: Optional[Union[dict, nn.Module]] = None, ) -> None: self.switch_epoch = switch_epoch if switch_data_preprocessor is None: switch_data_preprocessor = dict(type='BaseDataPreprocessor') if isinstance(switch_data_preprocessor, nn.Module): self.switch_data_preprocessor = switch_data_preprocessor elif isinstance(switch_data_preprocessor, dict): self.switch_data_preprocessor = MODELS.build( switch_data_preprocessor) else: raise TypeError('switch_data_preprocessor should be a `dict` or ' '`nn.Module` instance, but got ' f'{type(switch_data_preprocessor)}') self._has_switched = False def before_train_epoch(self, runner) -> None: epoch = runner.epoch model = runner.model # TODO: refactor after mmengine using model wrapper if is_model_wrapper(model): model = model.module if epoch >= self.switch_epoch and not self._has_switched: runner.logger.info( f'Switch data_preprocessor after epoch: {self.switch_epoch}!') model.data_preprocessor = self.switch_data_preprocessor.to( model.data_preprocessor.device) self._has_switched = True