mmdet配置文件到执行的过程

Posted 2022-02-05 东东就是我

1.读取配置文件，根据配置文件创建模型实例

1.1读取配置文件

model = build_detector(
        cfg.model,
        train_cfg=cfg.get('train_cfg'),
        test_cfg=cfg.get('test_cfg'))

1.2根据配置文件的type去寻找注册好的模型

def build_detector(cfg, train_cfg=None, test_cfg=None):
    """Build detector."""
    if train_cfg is not None or test_cfg is not None:
        warnings.warn(
            'train_cfg and test_cfg is deprecated, '
            'please specify them in model', UserWarning)
    assert cfg.get('train_cfg') is None or train_cfg is None, \\
        'train_cfg specified in both outer field and model field '
    assert cfg.get('test_cfg') is None or test_cfg is None, \\
        'test_cfg specified in both outer field and model field '
    return DETECTORS.build(
        cfg, default_args=dict(train_cfg=train_cfg, test_cfg=test_cfg))

    def build(self, *args, **kwargs):
        return self.build_func(*args, **kwargs, registry=self)

通过配置文件找到模型后，根据配置文件实例化模型类

def build_from_cfg(cfg, registry, default_args=None):
    """Build a module from config dict.

    Args:
        cfg (dict): Config dict. It should at least contain the key "type".
        registry (:obj:`Registry`): The registry to search the type from.
        default_args (dict, optional): Default initialization arguments.

    Returns:
        object: The constructed object.
    """
    if not isinstance(cfg, dict):
        raise TypeError(f'cfg must be a dict, but got type(cfg)')
    if 'type' not in cfg:
        if default_args is None or 'type' not in default_args:
            raise KeyError(
                '`cfg` or `default_args` must contain the key "type", '
                f'but got cfg\\ndefault_args')
    if not isinstance(registry, Registry):
        raise TypeError('registry must be an mmcv.Registry object, '
                        f'but got type(registry)')
    if not (isinstance(default_args, dict) or default_args is None):
        raise TypeError('default_args must be a dict or None, '
                        f'but got type(default_args)')

    args = cfg.copy()

    if default_args is not None:
        for name, value in default_args.items():
            args.setdefault(name, value)

    obj_type = args.pop('type')
    if isinstance(obj_type, str):
        obj_cls = registry.get(obj_type)
        if obj_cls is None:
            raise KeyError(
                f'obj_type is not in the registry.name registry')
    elif inspect.isclass(obj_type):
        obj_cls = obj_type
    else:
        raise TypeError(
            f'type must be a str or valid type, but got type(obj_type)')
    try:
        return obj_cls(**args)
    except Exception as e:
        # Normal TypeError does not print class name.
        raise type(e)(f'obj_cls.__name__: e')

2.模型实例注册（fasterrcnn为例）

2.1注册到detectors上

@DETECTORS.register_module()
class FasterRCNN(TwoStageDetector):
    """Implementation of `Faster R-CNN <https://arxiv.org/abs/1506.01497>`_"""

    def __init__(self,
                 backbone,
                 rpn_head,
                 roi_head,
                 train_cfg,
                 test_cfg,
                 neck=None,
                 pretrained=None,
                 init_cfg=None):
        super(FasterRCNN, self).__init__(
            backbone=backbone,
            neck=neck,
            rpn_head=rpn_head,
            roi_head=roi_head,
            train_cfg=train_cfg,
            test_cfg=test_cfg,
            pretrained=pretrained,
            init_cfg=init_cfg)

2.2detectors也就是models

import warnings

from mmcv.cnn import MODELS as MMCV_MODELS
from mmcv.utils import Registry

MODELS = Registry('models', parent=MMCV_MODELS)

BACKBONES = MODELS
NECKS = MODELS
ROI_EXTRACTORS = MODELS
SHARED_HEADS = MODELS
HEADS = MODELS
LOSSES = MODELS
DETECTORS = MODELS

2.3父类是TwoStageDetector

我们主要看train_step ，因为下面再说

 def train_step(self, data, optimizer):
        """The iteration step during training.

        This method defines an iteration step during training, except for the
        back propagation and optimizer updating, which are done in an optimizer
        hook. Note that in some complicated cases or models, the whole process
        including back propagation and optimizer updating is also defined in
        this method, such as GAN.

        Args:
            data (dict): The output of dataloader.
            optimizer (:obj:`torch.optim.Optimizer` | dict): The optimizer of
                runner is passed to ``train_step()``. This argument is unused
                and reserved.

        Returns:
            dict: It should contain at least 3 keys: ``loss``, ``log_vars``, \\
                ``num_samples``.

                - ``loss`` is a tensor for back propagation, which can be a
                  weighted sum of multiple losses.
                - ``log_vars`` contains all the variables to be sent to the
                  logger.
                - ``num_samples`` indicates the batch size (when the model is
                  DDP, it means the batch size on each GPU), which is used for
                  averaging the logs.
        """
        losses = self(**data)
        loss, log_vars = self._parse_losses(losses)

        outputs = dict(
            loss=loss, log_vars=log_vars, num_samples=len(data['img_metas']))

        return outputs

2.4 self（**data）

这个方法执行的是__cell__

所以 losses = self(**data) 就是执行了__call__（）
_call_impl 又执行了self.forward

__call__ : Callable[..., Any] = _call_impl

    def _call_impl(self, *input, **kwargs):
        forward_call = (self._slow_forward if torch._C._get_tracing_state() else self.forward)
        # If we don't have any hooks, we want to skip the rest of the logic in
        # this function, and just call forward.
        if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
                or _global_forward_hooks or _global_forward_pre_hooks):
            return forward_call(*input, **kwargs)
        # Do not call functions when jit is used
        full_backward_hooks, non_full_backward_hooks = [], []
        if self._backward_hooks or _global_backward_hooks:
            full_backward_hooks, non_full_backward_hooks = self._get_backward_hooks()
        if _global_forward_pre_hooks or self._forward_pre_hooks:
            for hook in itertools.chain(
                    _global_forward_pre_hooks.values(),
                    self._forward_pre_hooks.values()):
                result = hook(self, input)
                if result is not None:
                    if not isinstance(result, tuple):
                        result = (result,)
                    input = result

        bw_hook = None
        if full_backward_hooks:
            bw_hook = hooks.BackwardHook(self, full_backward_hooks)
            input = bw_hook.setup_input_hook(input)

        result = forward_call(*input, **kwargs)
        if _global_forward_hooks or self._forward_hooks:
            for hook in itertools.chain(
                    _global_forward_hooks.values(),
                    self._forward_hooks.values()):
                hook_result = hook(self, input, result)
                if hook_result is not None:
                    result = hook_result

        if bw_hook:
            result = bw_hook.setup_output_hook(result)

        # Handle the non-full backward hooks
        if non_full_backward_hooks:
            var = result
            while not isinstance(var, torch.Tensor):
                if isinstance(var, dict):
                    var = next((v for v in var.values() if isinstance(v, torch.Tensor)))
                else:
                    var = var[0]
            grad_fn = var.grad_fn
            if grad_fn is not None:
                for hook in non_full_backward_hooks:
                    wrapper = functools.partial(hook, self)
                    functools.update_wrapper(wrapper, hook)
                    grad_fn.register_hook(wrapper)
                self._maybe_warn_non_full_backward_hook(input, result, grad_fn)

        return result

forward 又执行了foraward_train

    def forward(self, img, img_metas, return_loss=True, **kwargs):
        """Calls either :func:`forward_train` or :func:`forward_test` depending
        on whether ``return_loss`` is ``True``.

        Note this setting will change the expected inputs. When
        ``return_loss=True``, img and img_meta are single-nested (i.e. Tensor
        and List[dict]), and when ``resturn_loss=False``, img and img_meta
        should be double nested (i.e.  List[Tensor], List[List[dict]]), with
        the outer list indicating test time augmentations.
        """
        if torch.onnx.is_in_onnx_export():
            assert len(img_metas) == 1
            return self.onnx_export(img[0], img_metas[0])

        if return_loss:
            return self.forward_train(img, img_metas, **kwargs)
        else:
            return self.forward_test(img, img_metas, **kwargs)

    def forward_train(self,
                      img,
                      img_metas,
                      gt_bboxes,
                      gt_labels,
                      gt_bboxes_ignore=None,
                      gt_masks=None,
                      proposals=None,
                      **kwargs):
        """
        Args:
            img (Tensor): of shape (N, C, H, W) encoding input images.
                Typically these should be mean centered and std scaled.

            img_metas (list[dict]): list of image info dict where each dict
                has: 'img_shape', 'scale_factor', 'flip', and may also contain
                'filename', 'ori_shape', 'pad_shape', and 'img_norm_cfg'.
                For details on the values of these keys see
                `mmdet/datasets/pipelines/formatting.py:Collect`.

            gt_bboxes (list[Tensor]): Ground truth bboxes for each image with
                shape (num_gts, 4) in [tl_x, tl_y, br_x, br_y] format.

            gt_labels (list[Tensor]): class indices corresponding to each box

            gt_bboxes_ignore (None | list[Tensor]): specify which bounding
                boxes can be ignored when computing the loss.

            gt_masks (None | Tensor) : true segmentation masks for each box
                used if the architecture supports a segmentation task.

            proposals : override rpn proposals with custom proposals. Use when
                `with_rpn` is False.

        Returns:
            dict[str, Tensor]: a dictionary of loss components
        """
        x = self.extract_feat(img)

        losses = dict()

        # RPN forward and loss
        if self.with_rpn:
            proposal_cfg = self.train_cfg.get('rpn_proposal',
                                              self.test_cfg.rpn)
            rpn_losses, proposal_list = self.rpn_head.forward_train(
                x,
                img_metas,
                gt_bboxes,
                gt_labels=None,
                gt_bboxes_ignore=gt_bboxes_ignore,
                proposal_cfg=proposal_cfg,
                **kwargs)
            losses.update(rpn_losses)
        else:
            proposal_list = proposals

        roi_losses = self.roi_head.forward_train(x, img_metas, proposal_list,
                                                 gt_bboxes, gt_labels,
                                                 gt_bboxes_ignore, gt_masks,
                                                 **kwargs)
        losses.update(roi_losses)

        return losses

3.模型训练

runner.run(data_loaders, cfg.workflow)

    def run(self, data_loaders, workflow, max_epochs=None, **kwargs):
        """Start running.

        Args:
            data_loaders (list[:obj:`DataLoader`]): Dataloaders for training
                and validation.
            workflow (list[tuple]): A list of (phase, epochs) to specify the
                running order and epochs. E.g, [('train', 2), ('val', 1)] means
                running 2 epochs for training and 1 epoch for validation,
                iteratively.
        """
        assert isinstance(data_loaders, list)
        assert mmcv.is_list_of(workflow, tuple)
        assert len(data_loaders) == len(workflow)
        if max_epochs is not None:
            warnings.warn(
                'setting max_epochs in run is deprecated, '
                'please set max_epochs in runner_config', DeprecationWarning)
            self._max_epochs = max_epochs

        assert self._max_epochs is not None, (
            'max_epochs must be specified during instantiation')

        for i, flow in enumerate(workflow):
            mode, epochs = flow
            if mode == 'train':
                self._max_iters = self._max_epochs * len(data_loaders[i])
                break

        work_dir = self.work_dir if self.work_dir is not None else 'NONE'
        self.logger.info('Start running, host: %s, work_dir: %s',
                         get_host_info(), work_dir)
        self.logger.info('Hooks will be executed in the following order:\\n%s',
                         self.get_hook_info())
        self.logger.info('workflow: %s, max: %d epochs', workflow,
                         self._max_epochs)
        self.call_hook('before_run')

        while self.epoch < self._max_epochs:
            for i, flow in enumerate(workflow):
                mode, epochs = flow
                if isinstance(mode, str):  # self.train()
                    if not hasattr(self, mode):
                        raise ValueError(
                            f'runner has no method named "mode" to run an '
                            'epoch')
                    epoch_runner = getattr(self, mode)
                else:
                    raise TypeError(
                        'mode in workflow must be a str, but got '.format(
                            type(mode)))

                for _ in range(epochs):
                    if mode == 'train' and self.epoch >= self._max_epochs:
                        break
                    epoch_runner(data_loaders[i], **kwargs)

        time.sleep(1)  # wait for some hooks like loggers to finish
        self.call_hook('after_run')

run 执行了epoch_runner .其中
epoch_runner = getattr(self, mode)
mode=train ，所以又执行了train

    def train(self, data_loader, **kwargs):
        self.model.train()
        self.mode = 'train'
        self.data_loader = data_loader
        self._max_iters = self._max_epochs * len(self.data_loader)
        self.call_hook('before_train_epoch')
        time.sleep(2)  # Prevent possible deadlock during epoch transition
        for i, data_batch in enumerate(self.data_loader):
            self._inner_iter = i
            self.call_hook('before_train_iter')
            self.run_iter(data_batch, train_mode=True, **kwargs)
            self.call_hook('after_train_iter')
            self._iter += 1

        self.call_hook('after_train_epoch')
        self._epoch += 1

run_iter 又执行了train_step ，所以上面我们关注模型中train_step方法

    def run_iter(self, data_batch, train_mode, **kwargs):
        if self.batch_processor is not None:
            outputs = self.batch_processor(
                self.model, data_batch, train_mode=train_mode, **kwargs)
        elif train_mode:
            outputs = self.model.train_step(data_batch, self.optimizer,
                                            **kwargs)
        else:
            outputs = self.model.val_step(data_batch, self.optimizer, **kwargs)
        if not isinstance(outputs, dict):
            raise TypeError('"batch_processor()" or "model.train_step()"'
                            'and "model.val_step()" must return a dict')
        if 'log_vars' in outputs:
            self.log_buffer.update(outputs['log_vars'], outputs['num_samples'])
        self.outputs = outputs