Add Swin Transformer model

mmcv_custom/__init__.py

+# -*- coding: utf-8 -*-
+
+from .checkpoint import load_checkpoint
+
+__all__ = ['load_checkpoint']

mmcv_custom/checkpoint.py

+# Copyright (c) Open-MMLab. All rights reserved.
+import io
+import os
+import os.path as osp
+import pkgutil
+import time
+import warnings
+from collections import OrderedDict
+from importlib import import_module
+from tempfile import TemporaryDirectory
+
+import torch
+import torchvision
+from torch.optim import Optimizer
+from torch.utils import model_zoo
+from torch.nn import functional as F
+
+import mmcv
+from mmcv.fileio import FileClient
+from mmcv.fileio import load as load_file
+from mmcv.parallel import is_module_wrapper
+from mmcv.utils import mkdir_or_exist
+from mmcv.runner import get_dist_info
+
+ENV_MMCV_HOME = 'MMCV_HOME'
+ENV_XDG_CACHE_HOME = 'XDG_CACHE_HOME'
+DEFAULT_CACHE_DIR = '~/.cache'
+
+
+def _get_mmcv_home():
+    mmcv_home = os.path.expanduser(
+        os.getenv(
+            ENV_MMCV_HOME,
+            os.path.join(
+                os.getenv(ENV_XDG_CACHE_HOME, DEFAULT_CACHE_DIR), 'mmcv')))
+
+    mkdir_or_exist(mmcv_home)
+    return mmcv_home
+
+
+def load_state_dict(module, state_dict, strict=False, logger=None):
+    """Load state_dict to a module.
+
+    This method is modified from :meth:`torch.nn.Module.load_state_dict`.
+    Default value for ``strict`` is set to ``False`` and the message for
+    param mismatch will be shown even if strict is False.
+
+    Args:
+        module (Module): Module that receives the state_dict.
+        state_dict (OrderedDict): Weights.
+        strict (bool): whether to strictly enforce that the keys
+            in :attr:`state_dict` match the keys returned by this module's
+            :meth:`~torch.nn.Module.state_dict` function. Default: ``False``.
+        logger (:obj:`logging.Logger`, optional): Logger to log the error
+            message. If not specified, print function will be used.
+    """
+    unexpected_keys = []
+    all_missing_keys = []
+    err_msg = []
+
+    metadata = getattr(state_dict, '_metadata', None)
+    state_dict = state_dict.copy()
+    if metadata is not None:
+        state_dict._metadata = metadata
+
+    # use _load_from_state_dict to enable checkpoint version control
+    def load(module, prefix=''):
+        # recursively check parallel module in case that the model has a
+        # complicated structure, e.g., nn.Module(nn.Module(DDP))
+        if is_module_wrapper(module):
+            module = module.module
+        local_metadata = {} if metadata is None else metadata.get(
+            prefix[:-1], {})
+        module._load_from_state_dict(state_dict, prefix, local_metadata, True,
+                                     all_missing_keys, unexpected_keys,
+                                     err_msg)
+        for name, child in module._modules.items():
+            if child is not None:
+                load(child, prefix + name + '.')
+
+    load(module)
+    load = None  # break load->load reference cycle
+
+    # ignore "num_batches_tracked" of BN layers
+    missing_keys = [
+        key for key in all_missing_keys if 'num_batches_tracked' not in key
+    ]
+
+    if unexpected_keys:
+        err_msg.append('unexpected key in source '
+                       f'state_dict: {", ".join(unexpected_keys)}\n')
+    if missing_keys:
+        err_msg.append(
+            f'missing keys in source state_dict: {", ".join(missing_keys)}\n')
+
+    rank, _ = get_dist_info()
+    if len(err_msg) > 0 and rank == 0:
+        err_msg.insert(
+            0, 'The model and loaded state dict do not match exactly\n')
+        err_msg = '\n'.join(err_msg)
+        if strict:
+            raise RuntimeError(err_msg)
+        elif logger is not None:
+            logger.warning(err_msg)
+        else:
+            print(err_msg)
+
+
+def load_url_dist(url, model_dir=None):
+    """In distributed setting, this function only download checkpoint at local
+    rank 0."""
+    rank, world_size = get_dist_info()
+    rank = int(os.environ.get('LOCAL_RANK', rank))
+    if rank == 0:
+        checkpoint = model_zoo.load_url(url, model_dir=model_dir)
+    if world_size > 1:
+        torch.distributed.barrier()
+        if rank > 0:
+            checkpoint = model_zoo.load_url(url, model_dir=model_dir)
+    return checkpoint
+
+
+def load_pavimodel_dist(model_path, map_location=None):
+    """In distributed setting, this function only download checkpoint at local
+    rank 0."""
+    try:
+        from pavi import modelcloud
+    except ImportError:
+        raise ImportError(
+            'Please install pavi to load checkpoint from modelcloud.')
+    rank, world_size = get_dist_info()
+    rank = int(os.environ.get('LOCAL_RANK', rank))
+    if rank == 0:
+        model = modelcloud.get(model_path)
+        with TemporaryDirectory() as tmp_dir:
+            downloaded_file = osp.join(tmp_dir, model.name)
+            model.download(downloaded_file)
+            checkpoint = torch.load(downloaded_file, map_location=map_location)
+    if world_size > 1:
+        torch.distributed.barrier()
+        if rank > 0:
+            model = modelcloud.get(model_path)
+            with TemporaryDirectory() as tmp_dir:
+                downloaded_file = osp.join(tmp_dir, model.name)
+                model.download(downloaded_file)
+                checkpoint = torch.load(
+                    downloaded_file, map_location=map_location)
+    return checkpoint
+
+
+def load_fileclient_dist(filename, backend, map_location):
+    """In distributed setting, this function only download checkpoint at local
+    rank 0."""
+    rank, world_size = get_dist_info()
+    rank = int(os.environ.get('LOCAL_RANK', rank))
+    allowed_backends = ['ceph']
+    if backend not in allowed_backends:
+        raise ValueError(f'Load from Backend {backend} is not supported.')
+    if rank == 0:
+        fileclient = FileClient(backend=backend)
+        buffer = io.BytesIO(fileclient.get(filename))
+        checkpoint = torch.load(buffer, map_location=map_location)
+    if world_size > 1:
+        torch.distributed.barrier()
+        if rank > 0:
+            fileclient = FileClient(backend=backend)
+            buffer = io.BytesIO(fileclient.get(filename))
+            checkpoint = torch.load(buffer, map_location=map_location)
+    return checkpoint
+
+
+def get_torchvision_models():
+    model_urls = dict()
+    for _, name, ispkg in pkgutil.walk_packages(torchvision.models.__path__):
+        if ispkg:
+            continue
+        _zoo = import_module(f'torchvision.models.{name}')
+        if hasattr(_zoo, 'model_urls'):
+            _urls = getattr(_zoo, 'model_urls')
+            model_urls.update(_urls)
+    return model_urls
+
+
+def get_external_models():
+    mmcv_home = _get_mmcv_home()
+    default_json_path = osp.join(mmcv.__path__[0], 'model_zoo/open_mmlab.json')
+    default_urls = load_file(default_json_path)
+    assert isinstance(default_urls, dict)
+    external_json_path = osp.join(mmcv_home, 'open_mmlab.json')
+    if osp.exists(external_json_path):
+        external_urls = load_file(external_json_path)
+        assert isinstance(external_urls, dict)
+        default_urls.update(external_urls)
+
+    return default_urls
+
+
+def get_mmcls_models():
+    mmcls_json_path = osp.join(mmcv.__path__[0], 'model_zoo/mmcls.json')
+    mmcls_urls = load_file(mmcls_json_path)
+
+    return mmcls_urls
+
+
+def get_deprecated_model_names():
+    deprecate_json_path = osp.join(mmcv.__path__[0],
+                                   'model_zoo/deprecated.json')
+    deprecate_urls = load_file(deprecate_json_path)
+    assert isinstance(deprecate_urls, dict)
+
+    return deprecate_urls
+
+
+def _process_mmcls_checkpoint(checkpoint):
+    state_dict = checkpoint['state_dict']
+    new_state_dict = OrderedDict()
+    for k, v in state_dict.items():
+        if k.startswith('backbone.'):
+            new_state_dict[k[9:]] = v
+    new_checkpoint = dict(state_dict=new_state_dict)
+
+    return new_checkpoint
+
+
+def _load_checkpoint(filename, map_location=None):
+    """Load checkpoint from somewhere (modelzoo, file, url).
+
+    Args:
+        filename (str): Accept local filepath, URL, ``torchvision://xxx``,
+            ``open-mmlab://xxx``. Please refer to ``docs/model_zoo.md`` for
+            details.
+        map_location (str | None): Same as :func:`torch.load`. Default: None.
+
+    Returns:
+        dict | OrderedDict: The loaded checkpoint. It can be either an
+            OrderedDict storing model weights or a dict containing other
+            information, which depends on the checkpoint.
+    """
+    if filename.startswith('modelzoo://'):
+        warnings.warn('The URL scheme of "modelzoo://" is deprecated, please '
+                      'use "torchvision://" instead')
+        model_urls = get_torchvision_models()
+        model_name = filename[11:]
+        checkpoint = load_url_dist(model_urls[model_name])
+    elif filename.startswith('torchvision://'):
+        model_urls = get_torchvision_models()
+        model_name = filename[14:]
+        checkpoint = load_url_dist(model_urls[model_name])
+    elif filename.startswith('open-mmlab://'):
+        model_urls = get_external_models()
+        model_name = filename[13:]
+        deprecated_urls = get_deprecated_model_names()
+        if model_name in deprecated_urls:
+            warnings.warn(f'open-mmlab://{model_name} is deprecated in favor '
+                          f'of open-mmlab://{deprecated_urls[model_name]}')
+            model_name = deprecated_urls[model_name]
+        model_url = model_urls[model_name]
+        # check if is url
+        if model_url.startswith(('http://', 'https://')):
+            checkpoint = load_url_dist(model_url)
+        else:
+            filename = osp.join(_get_mmcv_home(), model_url)
+            if not osp.isfile(filename):
+                raise IOError(f'{filename} is not a checkpoint file')
+            checkpoint = torch.load(filename, map_location=map_location)
+    elif filename.startswith('mmcls://'):
+        model_urls = get_mmcls_models()
+        model_name = filename[8:]
+        checkpoint = load_url_dist(model_urls[model_name])
+        checkpoint = _process_mmcls_checkpoint(checkpoint)
+    elif filename.startswith(('http://', 'https://')):
+        checkpoint = load_url_dist(filename)
+    elif filename.startswith('pavi://'):
+        model_path = filename[7:]
+        checkpoint = load_pavimodel_dist(model_path, map_location=map_location)
+    elif filename.startswith('s3://'):
+        checkpoint = load_fileclient_dist(
+            filename, backend='ceph', map_location=map_location)
+    else:
+        if not osp.isfile(filename):
+            raise IOError(f'{filename} is not a checkpoint file')
+        checkpoint = torch.load(filename, map_location=map_location)
+    return checkpoint
+
+
+def load_checkpoint(model,
+                    filename,
+                    map_location='cpu',
+                    strict=False,
+                    logger=None):
+    """Load checkpoint from a file or URI.
+
+    Args:
+        model (Module): Module to load checkpoint.
+        filename (str): Accept local filepath, URL, ``torchvision://xxx``,
+            ``open-mmlab://xxx``. Please refer to ``docs/model_zoo.md`` for
+            details.
+        map_location (str): Same as :func:`torch.load`.
+        strict (bool): Whether to allow different params for the model and
+            checkpoint.
+        logger (:mod:`logging.Logger` or None): The logger for error message.
+
+    Returns:
+        dict or OrderedDict: The loaded checkpoint.
+    """
+    checkpoint = _load_checkpoint(filename, map_location)
+    # OrderedDict is a subclass of dict
+    if not isinstance(checkpoint, dict):
+        raise RuntimeError(
+            f'No state_dict found in checkpoint file {filename}')
+    # get state_dict from checkpoint
+    if 'state_dict' in checkpoint:
+        state_dict = checkpoint['state_dict']
+    elif 'model' in checkpoint:
+        state_dict = checkpoint['model']
+    else:
+        state_dict = checkpoint
+    # strip prefix of state_dict
+    if list(state_dict.keys())[0].startswith('module.'):
+        state_dict = {k[7:]: v for k, v in state_dict.items()}
+
+    # reshape absolute position embedding
+    if state_dict.get('absolute_pos_embed') is not None:
+        absolute_pos_embed = state_dict['absolute_pos_embed']
+        N1, L, C1 = absolute_pos_embed.size()
+        N2, C2, H, W = model.absolute_pos_embed.size()
+        if N1 != N2 or C1 != C2 or L != H*W:
+            logger.warning("Error in loading absolute_pos_embed, pass")
+        else:
+            state_dict['absolute_pos_embed'] = absolute_pos_embed.view(N2, H, W, C2).permute(0, 3, 1, 2)
+
+    # interpolate position bias table if needed
+    relative_position_bias_table_keys = [k for k in state_dict.keys() if "relative_position_bias_table" in k]
+    for table_key in relative_position_bias_table_keys:
+        table_pretrained = state_dict[table_key]
+        table_current = model.state_dict()[table_key]
+        L1, nH1 = table_pretrained.size()
+        L2, nH2 = table_current.size()
+        if nH1 != nH2:
+            logger.warning(f"Error in loading {table_key}, pass")
+        else:
+            if L1 != L2:
+                S1 = int(L1 ** 0.5)
+                S2 = int(L2 ** 0.5)
+                table_pretrained_resized = F.interpolate(
+                     table_pretrained.permute(1, 0).view(1, nH1, S1, S1),
+                     size=(S2, S2), mode='bicubic')
+                state_dict[table_key] = table_pretrained_resized.view(nH2, L2).permute(1, 0)
+
+    # load state_dict
+    load_state_dict(model, state_dict, strict, logger)
+    return checkpoint
+
+
+def weights_to_cpu(state_dict):
+    """Copy a model state_dict to cpu.
+
+    Args:
+        state_dict (OrderedDict): Model weights on GPU.
+
+    Returns:
+        OrderedDict: Model weights on GPU.
+    """
+    state_dict_cpu = OrderedDict()
+    for key, val in state_dict.items():
+        state_dict_cpu[key] = val.cpu()
+    return state_dict_cpu
+
+
+def _save_to_state_dict(module, destination, prefix, keep_vars):
+    """Saves module state to `destination` dictionary.
+
+    This method is modified from :meth:`torch.nn.Module._save_to_state_dict`.
+
+    Args:
+        module (nn.Module): The module to generate state_dict.
+        destination (dict): A dict where state will be stored.
+        prefix (str): The prefix for parameters and buffers used in this
+            module.
+    """
+    for name, param in module._parameters.items():
+        if param is not None:
+            destination[prefix + name] = param if keep_vars else param.detach()
+    for name, buf in module._buffers.items():
+        # remove check of _non_persistent_buffers_set to allow nn.BatchNorm2d
+        if buf is not None:
+            destination[prefix + name] = buf if keep_vars else buf.detach()
+
+
+def get_state_dict(module, destination=None, prefix='', keep_vars=False):
+    """Returns a dictionary containing a whole state of the module.
+
+    Both parameters and persistent buffers (e.g. running averages) are
+    included. Keys are corresponding parameter and buffer names.
+
+    This method is modified from :meth:`torch.nn.Module.state_dict` to
+    recursively check parallel module in case that the model has a complicated
+    structure, e.g., nn.Module(nn.Module(DDP)).
+
+    Args:
+        module (nn.Module): The module to generate state_dict.
+        destination (OrderedDict): Returned dict for the state of the
+            module.
+        prefix (str): Prefix of the key.
+        keep_vars (bool): Whether to keep the variable property of the
+            parameters. Default: False.
+
+    Returns:
+        dict: A dictionary containing a whole state of the module.
+    """
+    # recursively check parallel module in case that the model has a
+    # complicated structure, e.g., nn.Module(nn.Module(DDP))
+    if is_module_wrapper(module):
+        module = module.module
+
+    # below is the same as torch.nn.Module.state_dict()
+    if destination is None:
+        destination = OrderedDict()
+        destination._metadata = OrderedDict()
+    destination._metadata[prefix[:-1]] = local_metadata = dict(
+        version=module._version)
+    _save_to_state_dict(module, destination, prefix, keep_vars)
+    for name, child in module._modules.items():
+        if child is not None:
+            get_state_dict(
+                child, destination, prefix + name + '.', keep_vars=keep_vars)
+    for hook in module._state_dict_hooks.values():
+        hook_result = hook(module, destination, prefix, local_metadata)
+        if hook_result is not None:
+            destination = hook_result
+    return destination
+
+
+def save_checkpoint(model, filename, optimizer=None, meta=None):
+    """Save checkpoint to file.
+
+    The checkpoint will have 3 fields: ``meta``, ``state_dict`` and
+    ``optimizer``. By default ``meta`` will contain version and time info.
+
+    Args:
+        model (Module): Module whose params are to be saved.
+        filename (str): Checkpoint filename.
+        optimizer (:obj:`Optimizer`, optional): Optimizer to be saved.
+        meta (dict, optional): Metadata to be saved in checkpoint.
+    """
+    if meta is None:
+        meta = {}
+    elif not isinstance(meta, dict):
+        raise TypeError(f'meta must be a dict or None, but got {type(meta)}')
+    meta.update(mmcv_version=mmcv.__version__, time=time.asctime())
+
+    if is_module_wrapper(model):
+        model = model.module
+
+    if hasattr(model, 'CLASSES') and model.CLASSES is not None:
+        # save class name to the meta
+        meta.update(CLASSES=model.CLASSES)
+
+    checkpoint = {
+        'meta': meta,
+        'state_dict': weights_to_cpu(get_state_dict(model))
+    }
+    # save optimizer state dict in the checkpoint
+    if isinstance(optimizer, Optimizer):
+        checkpoint['optimizer'] = optimizer.state_dict()
+    elif isinstance(optimizer, dict):
+        checkpoint['optimizer'] = {}
+        for name, optim in optimizer.items():
+            checkpoint['optimizer'][name] = optim.state_dict()
+
+    if filename.startswith('pavi://'):
+        try:
+            from pavi import modelcloud
+            from pavi.exception import NodeNotFoundError
+        except ImportError:
+            raise ImportError(
+                'Please install pavi to load checkpoint from modelcloud.')
+        model_path = filename[7:]
+        root = modelcloud.Folder()
+        model_dir, model_name = osp.split(model_path)
+        try:
+            model = modelcloud.get(model_dir)
+        except NodeNotFoundError:
+            model = root.create_training_model(model_dir)
+        with TemporaryDirectory() as tmp_dir:
+            checkpoint_file = osp.join(tmp_dir, model_name)
+            with open(checkpoint_file, 'wb') as f:
+                torch.save(checkpoint, f)
+                f.flush()
+            model.create_file(checkpoint_file, name=model_name)
+    else:
+        mmcv.mkdir_or_exist(osp.dirname(filename))
+        # immediately flush buffer
+        with open(filename, 'wb') as f:
+            torch.save(checkpoint, f)
+            f.flush()

mmdet/models/backbones/__init__.py

@@ -10,9 +10,10 @@ from .resnet import ResNet, ResNetV1d
 from .resnext import ResNeXt
 from .ssd_vgg import SSDVGG
 from .trident_resnet import TridentResNet
+from .swin_transformer import SwinTransformer
 
 __all__ = [
     'RegNet', 'ResNet', 'ResNetV1d', 'ResNeXt', 'SSDVGG', 'HRNet', 'Res2Net',
     'HourglassNet', 'DetectoRS_ResNet', 'DetectoRS_ResNeXt', 'Darknet',
-    'ResNeSt', 'TridentResNet'
+    'ResNeSt', 'TridentResNet', 'SwinTransformer'
 ]

mmdet/models/backbones/swin_transformer.py

+# --------------------------------------------------------
+# Swin Transformer
+# Copyright (c) 2021 Microsoft
+# Licensed under The MIT License [see LICENSE for details]
+# Written by Ze Liu, Yutong Lin, Yixuan Wei
+# --------------------------------------------------------
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+import torch.utils.checkpoint as checkpoint
+import numpy as np
+from timm.models.layers import DropPath, to_2tuple, trunc_normal_
+
+from mmcv_custom import load_checkpoint
+from mmdet.utils import get_root_logger
+from ..builder import BACKBONES
+
+
+class Mlp(nn.Module):
+    """ Multilayer perceptron."""
+
+    def __init__(self, in_features, hidden_features=None, out_features=None, act_layer=nn.GELU, drop=0.):
+        super().__init__()
+        out_features = out_features or in_features
+        hidden_features = hidden_features or in_features
+        self.fc1 = nn.Linear(in_features, hidden_features)
+        self.act = act_layer()
+        self.fc2 = nn.Linear(hidden_features, out_features)
+        self.drop = nn.Dropout(drop)
+
+    def forward(self, x):
+        x = self.fc1(x)
+        x = self.act(x)
+        x = self.drop(x)
+        x = self.fc2(x)
+        x = self.drop(x)
+        return x
+
+
+def window_partition(x, window_size):
+    """
+    Args:
+        x: (B, H, W, C)
+        window_size (int): window size
+
+    Returns:
+        windows: (num_windows*B, window_size, window_size, C)
+    """
+    B, H, W, C = x.shape
+    x = x.view(B, H // window_size, window_size, W // window_size, window_size, C)
+    windows = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(-1, window_size, window_size, C)
+    return windows
+
+
+def window_reverse(windows, window_size, H, W):
+    """
+    Args:
+        windows: (num_windows*B, window_size, window_size, C)
+        window_size (int): Window size
+        H (int): Height of image
+        W (int): Width of image
+
+    Returns:
+        x: (B, H, W, C)
+    """
+    B = int(windows.shape[0] / (H * W / window_size / window_size))
+    x = windows.view(B, H // window_size, W // window_size, window_size, window_size, -1)
+    x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1)
+    return x
+
+
+class WindowAttention(nn.Module):
+    """ Window based multi-head self attention (W-MSA) module with relative position bias.
+    It supports both of shifted and non-shifted window.
+
+    Args:
+        dim (int): Number of input channels.
+        window_size (tuple[int]): The height and width of the window.
+        num_heads (int): Number of attention heads.
+        qkv_bias (bool, optional):  If True, add a learnable bias to query, key, value. Default: True
+        qk_scale (float | None, optional): Override default qk scale of head_dim ** -0.5 if set
+        attn_drop (float, optional): Dropout ratio of attention weight. Default: 0.0
+        proj_drop (float, optional): Dropout ratio of output. Default: 0.0
+    """
+
+    def __init__(self, dim, window_size, num_heads, qkv_bias=True, qk_scale=None, attn_drop=0., proj_drop=0.):
+
+        super().__init__()
+        self.dim = dim
+        self.window_size = window_size  # Wh, Ww
+        self.num_heads = num_heads
+        head_dim = dim // num_heads
+        self.scale = qk_scale or head_dim ** -0.5
+
+        # define a parameter table of relative position bias
+        self.relative_position_bias_table = nn.Parameter(
+            torch.zeros((2 * window_size[0] - 1) * (2 * window_size[1] - 1), num_heads))  # 2*Wh-1 * 2*Ww-1, nH
+
+        # get pair-wise relative position index for each token inside the window
+        coords_h = torch.arange(self.window_size[0])
+        coords_w = torch.arange(self.window_size[1])
+        coords = torch.stack(torch.meshgrid([coords_h, coords_w]))  # 2, Wh, Ww
+        coords_flatten = torch.flatten(coords, 1)  # 2, Wh*Ww
+        relative_coords = coords_flatten[:, :, None] - coords_flatten[:, None, :]  # 2, Wh*Ww, Wh*Ww
+        relative_coords = relative_coords.permute(1, 2, 0).contiguous()  # Wh*Ww, Wh*Ww, 2
+        relative_coords[:, :, 0] += self.window_size[0] - 1  # shift to start from 0
+        relative_coords[:, :, 1] += self.window_size[1] - 1
+        relative_coords[:, :, 0] *= 2 * self.window_size[1] - 1
+        relative_position_index = relative_coords.sum(-1)  # Wh*Ww, Wh*Ww
+        self.register_buffer("relative_position_index", relative_position_index)
+
+        self.qkv = nn.Linear(dim, dim * 3, bias=qkv_bias)
+        self.attn_drop = nn.Dropout(attn_drop)
+        self.proj = nn.Linear(dim, dim)
+        self.proj_drop = nn.Dropout(proj_drop)
+
+        trunc_normal_(self.relative_position_bias_table, std=.02)
+        self.softmax = nn.Softmax(dim=-1)
+
+    def forward(self, x, mask=None):
+        """ Forward function.
+
+        Args:
+            x: input features with shape of (num_windows*B, N, C)
+            mask: (0/-inf) mask with shape of (num_windows, Wh*Ww, Wh*Ww) or None
+        """
+        B_, N, C = x.shape
+        qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4)
+        q, k, v = qkv[0], qkv[1], qkv[2]  # make torchscript happy (cannot use tensor as tuple)
+
+        q = q * self.scale
+        attn = (q @ k.transpose(-2, -1))
+
+        relative_position_bias = self.relative_position_bias_table[self.relative_position_index.view(-1)].view(
+            self.window_size[0] * self.window_size[1], self.window_size[0] * self.window_size[1], -1)  # Wh*Ww,Wh*Ww,nH
+        relative_position_bias = relative_position_bias.permute(2, 0, 1).contiguous()  # nH, Wh*Ww, Wh*Ww
+        attn = attn + relative_position_bias.unsqueeze(0)
+
+        if mask is not None:
+            nW = mask.shape[0]
+            attn = attn.view(B_ // nW, nW, self.num_heads, N, N) + mask.unsqueeze(1).unsqueeze(0)
+            attn = attn.view(-1, self.num_heads, N, N)
+            attn = self.softmax(attn)
+        else:
+            attn = self.softmax(attn)
+
+        attn = self.attn_drop(attn)
+
+        x = (attn @ v).transpose(1, 2).reshape(B_, N, C)
+        x = self.proj(x)
+        x = self.proj_drop(x)
+        return x
+
+
+class SwinTransformerBlock(nn.Module):
+    """ Swin Transformer Block.
+
+    Args:
+        dim (int): Number of input channels.
+        num_heads (int): Number of attention heads.
+        window_size (int): Window size.
+        shift_size (int): Shift size for SW-MSA.
+        mlp_ratio (float): Ratio of mlp hidden dim to embedding dim.
+        qkv_bias (bool, optional): If True, add a learnable bias to query, key, value. Default: True
+        qk_scale (float | None, optional): Override default qk scale of head_dim ** -0.5 if set.
+        drop (float, optional): Dropout rate. Default: 0.0
+        attn_drop (float, optional): Attention dropout rate. Default: 0.0
+        drop_path (float, optional): Stochastic depth rate. Default: 0.0
+        act_layer (nn.Module, optional): Activation layer. Default: nn.GELU
+        norm_layer (nn.Module, optional): Normalization layer.  Default: nn.LayerNorm
+    """
+
+    def __init__(self, dim, num_heads, window_size=7, shift_size=0,
+                 mlp_ratio=4., qkv_bias=True, qk_scale=None, drop=0., attn_drop=0., drop_path=0.,
+                 act_layer=nn.GELU, norm_layer=nn.LayerNorm):
+        super().__init__()
+        self.dim = dim
+        self.num_heads = num_heads
+        self.window_size = window_size
+        self.shift_size = shift_size
+        self.mlp_ratio = mlp_ratio
+        assert 0 <= self.shift_size < self.window_size, "shift_size must in 0-window_size"
+
+        self.norm1 = norm_layer(dim)
+        self.attn = WindowAttention(
+            dim, window_size=to_2tuple(self.window_size), num_heads=num_heads,
+            qkv_bias=qkv_bias, qk_scale=qk_scale, attn_drop=attn_drop, proj_drop=drop)
+
+        self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()
+        self.norm2 = norm_layer(dim)
+        mlp_hidden_dim = int(dim * mlp_ratio)
+        self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop)
+
+        self.H = None
+        self.W = None
+
+    def forward(self, x, mask_matrix):
+        """ Forward function.
+
+        Args:
+            x: Input feature, tensor size (B, H*W, C).
+            H, W: Spatial resolution of the input feature.
+            mask_matrix: Attention mask for cyclic shift.
+        """
+        B, L, C = x.shape
+        H, W = self.H, self.W
+        assert L == H * W, "input feature has wrong size"
+
+        shortcut = x
+        x = self.norm1(x)
+        x = x.view(B, H, W, C)
+
+        # pad feature maps to multiples of window size
+        pad_l = pad_t = 0
+        pad_r = (self.window_size - W % self.window_size) % self.window_size
+        pad_b = (self.window_size - H % self.window_size) % self.window_size
+        x = F.pad(x, (0, 0, pad_l, pad_r, pad_t, pad_b))
+        _, Hp, Wp, _ = x.shape
+
+        # cyclic shift
+        if self.shift_size > 0:
+            shifted_x = torch.roll(x, shifts=(-self.shift_size, -self.shift_size), dims=(1, 2))
+            attn_mask = mask_matrix
+        else:
+            shifted_x = x
+            attn_mask = None
+
+        # partition windows
+        x_windows = window_partition(shifted_x, self.window_size)  # nW*B, window_size, window_size, C
+        x_windows = x_windows.view(-1, self.window_size * self.window_size, C)  # nW*B, window_size*window_size, C
+
+        # W-MSA/SW-MSA
+        attn_windows = self.attn(x_windows, mask=attn_mask)  # nW*B, window_size*window_size, C
+
+        # merge windows
+        attn_windows = attn_windows.view(-1, self.window_size, self.window_size, C)
+        shifted_x = window_reverse(attn_windows, self.window_size, Hp, Wp)  # B H' W' C
+
+        # reverse cyclic shift
+        if self.shift_size > 0:
+            x = torch.roll(shifted_x, shifts=(self.shift_size, self.shift_size), dims=(1, 2))
+        else:
+            x = shifted_x
+
+        if pad_r > 0 or pad_b > 0:
+            x = x[:, :H, :W, :].contiguous()
+
+        x = x.view(B, H * W, C)
+
+        # FFN
+        x = shortcut + self.drop_path(x)
+        x = x + self.drop_path(self.mlp(self.norm2(x)))
+
+        return x
+
+
+class PatchMerging(nn.Module):
+    """ Patch Merging Layer
+
+    Args:
+        dim (int): Number of input channels.
+        norm_layer (nn.Module, optional): Normalization layer.  Default: nn.LayerNorm
+    """
+    def __init__(self, dim, norm_layer=nn.LayerNorm):
+        super().__init__()
+        self.dim = dim
+        self.reduction = nn.Linear(4 * dim, 2 * dim, bias=False)
+        self.norm = norm_layer(4 * dim)
+
+    def forward(self, x, H, W):
+        """ Forward function.
+
+        Args:
+            x: Input feature, tensor size (B, H*W, C).
+            H, W: Spatial resolution of the input feature.
+        """
+        B, L, C = x.shape
+        assert L == H * W, "input feature has wrong size"
+
+        x = x.view(B, H, W, C)
+
+        # padding
+        pad_input = (H % 2 == 1) or (W % 2 == 1)
+        if pad_input:
+            x = F.pad(x, (0, 0, 0, W % 2, 0, H % 2))
+
+        x0 = x[:, 0::2, 0::2, :]  # B H/2 W/2 C
+        x1 = x[:, 1::2, 0::2, :]  # B H/2 W/2 C
+        x2 = x[:, 0::2, 1::2, :]  # B H/2 W/2 C
+        x3 = x[:, 1::2, 1::2, :]  # B H/2 W/2 C
+        x = torch.cat([x0, x1, x2, x3], -1)  # B H/2 W/2 4*C
+        x = x.view(B, -1, 4 * C)  # B H/2*W/2 4*C
+
+        x = self.norm(x)
+        x = self.reduction(x)
+
+        return x
+
+
+class BasicLayer(nn.Module):
+    """ A basic Swin Transformer layer for one stage.
+
+    Args:
+        dim (int): Number of feature channels
+        depth (int): Depths of this stage.
+        num_heads (int): Number of attention head.
+        window_size (int): Local window size. Default: 7.
+        mlp_ratio (float): Ratio of mlp hidden dim to embedding dim. Default: 4.
+        qkv_bias (bool, optional): If True, add a learnable bias to query, key, value. Default: True
+        qk_scale (float | None, optional): Override default qk scale of head_dim ** -0.5 if set.
+        drop (float, optional): Dropout rate. Default: 0.0
+        attn_drop (float, optional): Attention dropout rate. Default: 0.0
+        drop_path (float | tuple[float], optional): Stochastic depth rate. Default: 0.0
+        norm_layer (nn.Module, optional): Normalization layer. Default: nn.LayerNorm
+        downsample (nn.Module | None, optional): Downsample layer at the end of the layer. Default: None
+        use_checkpoint (bool): Whether to use checkpointing to save memory. Default: False.
+    """
+
+    def __init__(self,
+                 dim,
+                 depth,
+                 num_heads,
+                 window_size=7,
+                 mlp_ratio=4.,
+                 qkv_bias=True,
+                 qk_scale=None,
+                 drop=0.,
+                 attn_drop=0.,
+                 drop_path=0.,
+                 norm_layer=nn.LayerNorm,
+                 downsample=None,
+                 use_checkpoint=False):
+        super().__init__()
+        self.window_size = window_size
+        self.shift_size = window_size // 2
+        self.depth = depth
+        self.use_checkpoint = use_checkpoint
+
+        # build blocks
+        self.blocks = nn.ModuleList([
+            SwinTransformerBlock(
+                dim=dim,
+                num_heads=num_heads,
+                window_size=window_size,
+                shift_size=0 if (i % 2 == 0) else window_size // 2,
+                mlp_ratio=mlp_ratio,
+                qkv_bias=qkv_bias,
+                qk_scale=qk_scale,
+                drop=drop,
+                attn_drop=attn_drop,
+                drop_path=drop_path[i] if isinstance(drop_path, list) else drop_path,
+                norm_layer=norm_layer)
+            for i in range(depth)])
+
+        # patch merging layer
+        if downsample is not None:
+            self.downsample = downsample(dim=dim, norm_layer=norm_layer)
+        else:
+            self.downsample = None
+
+    def forward(self, x, H, W):
+        """ Forward function.
+
+        Args:
+            x: Input feature, tensor size (B, H*W, C).
+            H, W: Spatial resolution of the input feature.
+        """
+
+        # calculate attention mask for SW-MSA
+        Hp = int(np.ceil(H / self.window_size)) * self.window_size
+        Wp = int(np.ceil(W / self.window_size)) * self.window_size
+        img_mask = torch.zeros((1, Hp, Wp, 1), device=x.device)  # 1 Hp Wp 1
+        h_slices = (slice(0, -self.window_size),
+                    slice(-self.window_size, -self.shift_size),
+                    slice(-self.shift_size, None))
+        w_slices = (slice(0, -self.window_size),
+                    slice(-self.window_size, -self.shift_size),
+                    slice(-self.shift_size, None))
+        cnt = 0
+        for h in h_slices:
+            for w in w_slices:
+                img_mask[:, h, w, :] = cnt
+                cnt += 1
+
+        mask_windows = window_partition(img_mask, self.window_size)  # nW, window_size, window_size, 1
+        mask_windows = mask_windows.view(-1, self.window_size * self.window_size)
+        attn_mask = mask_windows.unsqueeze(1) - mask_windows.unsqueeze(2)
+        attn_mask = attn_mask.masked_fill(attn_mask != 0, float(-100.0)).masked_fill(attn_mask == 0, float(0.0))
+
+        for blk in self.blocks:
+            blk.H, blk.W = H, W
+            if self.use_checkpoint:
+                x = checkpoint.checkpoint(blk, x, attn_mask)
+            else:
+                x = blk(x, attn_mask)
+        if self.downsample is not None:
+            x_down = self.downsample(x, H, W)
+            Wh, Ww = (H + 1) // 2, (W + 1) // 2
+            return x, H, W, x_down, Wh, Ww
+        else:
+            return x, H, W, x, H, W
+
+
+class PatchEmbed(nn.Module):
+    """ Image to Patch Embedding
+
+    Args:
+        patch_size (int): Patch token size. Default: 4.
+        in_chans (int): Number of input image channels. Default: 3.
+        embed_dim (int): Number of linear projection output channels. Default: 96.
+        norm_layer (nn.Module, optional): Normalization layer. Default: None
+    """
+
+    def __init__(self, patch_size=4, in_chans=3, embed_dim=96, norm_layer=None):
+        super().__init__()
+        patch_size = to_2tuple(patch_size)
+        self.patch_size = patch_size
+
+        self.in_chans = in_chans
+        self.embed_dim = embed_dim
+
+        self.proj = nn.Conv2d(in_chans, embed_dim, kernel_size=patch_size, stride=patch_size)
+        if norm_layer is not None:
+            self.norm = norm_layer(embed_dim)
+        else:
+            self.norm = None
+
+    def forward(self, x):
+        """Forward function."""
+        # padding
+        _, _, H, W = x.size()
+        if W % self.patch_size[1] != 0:
+            x = F.pad(x, (0, self.patch_size[1] - W % self.patch_size[1]))
+        if H % self.patch_size[0] != 0:
+            x = F.pad(x, (0, 0, 0, self.patch_size[0] - H % self.patch_size[0]))
+
+        x = self.proj(x)  # B C Wh Ww
+        if self.norm is not None:
+            Wh, Ww = x.size(2), x.size(3)
+            x = x.flatten(2).transpose(1, 2)
+            x = self.norm(x)
+            x = x.transpose(1, 2).view(-1, self.embed_dim, Wh, Ww)
+
+        return x
+
+
+@BACKBONES.register_module()
+class SwinTransformer(nn.Module):
+    """ Swin Transformer backbone.
+        A PyTorch impl of : `Swin Transformer: Hierarchical Vision Transformer using Shifted Windows`  -
+          https://arxiv.org/pdf/2103.14030
+
+    Args:
+        pretrain_img_size (int): Input image size for training the pretrained model,
+            used in absolute postion embedding. Default 224.
+        patch_size (int | tuple(int)): Patch size. Default: 4.
+        in_chans (int): Number of input image channels. Default: 3.
+        embed_dim (int): Number of linear projection output channels. Default: 96.
+        depths (tuple[int]): Depths of each Swin Transformer stage.
+        num_heads (tuple[int]): Number of attention head of each stage.
+        window_size (int): Window size. Default: 7.
+        mlp_ratio (float): Ratio of mlp hidden dim to embedding dim. Default: 4.
+        qkv_bias (bool): If True, add a learnable bias to query, key, value. Default: True
+        qk_scale (float): Override default qk scale of head_dim ** -0.5 if set.
+        drop_rate (float): Dropout rate.
+        attn_drop_rate (float): Attention dropout rate. Default: 0.
+        drop_path_rate (float): Stochastic depth rate. Default: 0.2.
+        norm_layer (nn.Module): Normalization layer. Default: nn.LayerNorm.
+        ape (bool): If True, add absolute position embedding to the patch embedding. Default: False.
+        patch_norm (bool): If True, add normalization after patch embedding. Default: True.
+        out_indices (Sequence[int]): Output from which stages.
+        frozen_stages (int): Stages to be frozen (stop grad and set eval mode).
+            -1 means not freezing any parameters.
+        use_checkpoint (bool): Whether to use checkpointing to save memory. Default: False.
+    """
+
+    def __init__(self,
+                 pretrain_img_size=224,
+                 patch_size=4,
+                 in_chans=3,
+                 embed_dim=96,
+                 depths=[2, 2, 6, 2],
+                 num_heads=[3, 6, 12, 24],
+                 window_size=7,
+                 mlp_ratio=4.,
+                 qkv_bias=True,
+                 qk_scale=None,
+                 drop_rate=0.,
+                 attn_drop_rate=0.,
+                 drop_path_rate=0.2,
+                 norm_layer=nn.LayerNorm,
+                 ape=False,
+                 patch_norm=True,
+                 out_indices=(0, 1, 2, 3),
+                 frozen_stages=-1,
+                 use_checkpoint=False):
+        super().__init__()
+
+        self.pretrain_img_size = pretrain_img_size
+        self.num_layers = len(depths)
+        self.embed_dim = embed_dim
+        self.ape = ape
+        self.patch_norm = patch_norm
+        self.out_indices = out_indices
+        self.frozen_stages = frozen_stages
+
+        # split image into non-overlapping patches
+        self.patch_embed = PatchEmbed(
+            patch_size=patch_size, in_chans=in_chans, embed_dim=embed_dim,
+            norm_layer=norm_layer if self.patch_norm else None)
+
+        # absolute position embedding
+        if self.ape:
+            pretrain_img_size = to_2tuple(pretrain_img_size)
+            patch_size = to_2tuple(patch_size)
+            patches_resolution = [pretrain_img_size[0] // patch_size[0], pretrain_img_size[1] // patch_size[1]]
+
+            self.absolute_pos_embed = nn.Parameter(torch.zeros(1, embed_dim, patches_resolution[0], patches_resolution[1]))
+            trunc_normal_(self.absolute_pos_embed, std=.02)
+
+        self.pos_drop = nn.Dropout(p=drop_rate)
+
+        # stochastic depth
+        dpr = [x.item() for x in torch.linspace(0, drop_path_rate, sum(depths))]  # stochastic depth decay rule
+
+        # build layers
+        self.layers = nn.ModuleList()
+        for i_layer in range(self.num_layers):
+            layer = BasicLayer(
+                dim=int(embed_dim * 2 ** i_layer),
+                depth=depths[i_layer],
+                num_heads=num_heads[i_layer],
+                window_size=window_size,
+                mlp_ratio=mlp_ratio,
+                qkv_bias=qkv_bias,
+                qk_scale=qk_scale,
+                drop=drop_rate,
+                attn_drop=attn_drop_rate,
+                drop_path=dpr[sum(depths[:i_layer]):sum(depths[:i_layer + 1])],
+                norm_layer=norm_layer,
+                downsample=PatchMerging if (i_layer < self.num_layers - 1) else None,
+                use_checkpoint=use_checkpoint)
+            self.layers.append(layer)
+
+        num_features = [int(embed_dim * 2 ** i) for i in range(self.num_layers)]
+        self.num_features = num_features
+
+        # add a norm layer for each output
+        for i_layer in out_indices:
+            layer = norm_layer(num_features[i_layer])
+            layer_name = f'norm{i_layer}'
+            self.add_module(layer_name, layer)
+
+        self._freeze_stages()
+
+    def _freeze_stages(self):
+        if self.frozen_stages >= 0:
+            self.patch_embed.eval()
+            for param in self.patch_embed.parameters():
+                param.requires_grad = False
+
+        if self.frozen_stages >= 1 and self.ape:
+            self.absolute_pos_embed.requires_grad = False
+
+        if self.frozen_stages >= 2:
+            self.pos_drop.eval()
+            for i in range(0, self.frozen_stages - 1):
+                m = self.layers[i]
+                m.eval()
+                for param in m.parameters():
+                    param.requires_grad = False
+
+    def init_weights(self, pretrained=None):
+        """Initialize the weights in backbone.
+
+        Args:
+            pretrained (str, optional): Path to pre-trained weights.
+                Defaults to None.
+        """
+
+        def _init_weights(m):
+            if isinstance(m, nn.Linear):
+                trunc_normal_(m.weight, std=.02)
+                if isinstance(m, nn.Linear) and m.bias is not None:
+                    nn.init.constant_(m.bias, 0)
+            elif isinstance(m, nn.LayerNorm):
+                nn.init.constant_(m.bias, 0)
+                nn.init.constant_(m.weight, 1.0)
+
+        if isinstance(pretrained, str):
+            self.apply(_init_weights)
+            logger = get_root_logger()
+            load_checkpoint(self, pretrained, strict=False, logger=logger)
+        elif pretrained is None:
+            self.apply(_init_weights)
+        else:
+            raise TypeError('pretrained must be a str or None')
+
+    def forward(self, x):
+        """Forward function."""
+        x = self.patch_embed(x)
+
+        Wh, Ww = x.size(2), x.size(3)
+        if self.ape:
+            # interpolate the position embedding to the corresponding size
+            absolute_pos_embed = F.interpolate(self.absolute_pos_embed, size=(Wh, Ww), mode='bicubic')
+            x = (x + absolute_pos_embed).flatten(2).transpose(1, 2)  # B Wh*Ww C
+        else:
+            x = x.flatten(2).transpose(1, 2)
+        x = self.pos_drop(x)
+
+        outs = []
+        for i in range(self.num_layers):
+            layer = self.layers[i]
+            x_out, H, W, x, Wh, Ww = layer(x, Wh, Ww)
+
+            if i in self.out_indices:
+                norm_layer = getattr(self, f'norm{i}')
+                x_out = norm_layer(x_out)
+
+                out = x_out.view(-1, H, W, self.num_features[i]).permute(0, 3, 1, 2).contiguous()
+                outs.append(out)
+
+        return tuple(outs)
+
+    def train(self, mode=True):
+        """Convert the model into training mode while keep layers freezed."""
+        super(SwinTransformer, self).train(mode)
+        self._freeze_stages()