# Copyright 2018-2023 OpenMMLab. All rights reserved.
# Reference: https://github.com/ViTAE-Transformer/ViTDet/blob/main/mmdet/models/backbones/vit.py
import math
from functools import partial

import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.utils.checkpoint as checkpoint
from mmcv.cnn import build_norm_layer, constant_init, kaiming_init
from mmcv.runner import get_dist_info
from mmdet.utils import get_root_logger
from timm.models.layers import drop_path, to_2tuple, trunc_normal_
from torch.nn.modules.batchnorm import _BatchNorm

from easycv.utils.checkpoint import load_checkpoint
from ..registry import BACKBONES
from ..utils import build_conv_layer


class DropPath(nn.Module):
    """Drop paths (Stochastic Depth) per sample  (when applied in main path of residual blocks).
    """

    def __init__(self, drop_prob=None):
        super(DropPath, self).__init__()
        self.drop_prob = drop_prob

    def forward(self, x):
        return drop_path(x, self.drop_prob, self.training)

    def extra_repr(self):
        return 'p={}'.format(self.drop_prob)


class Mlp(nn.Module):

    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)
        # commit this for the orignal BERT implement
        x = self.fc2(x)
        x = self.drop(x)
        return x


class BasicBlock(nn.Module):
    expansion = 1

    def __init__(self,
                 inplanes,
                 planes,
                 stride=1,
                 dilation=1,
                 conv_cfg=None,
                 norm_cfg=dict(type='BN')):
        super(BasicBlock, self).__init__()

        self.norm1_name, norm1 = build_norm_layer(norm_cfg, planes, postfix=1)
        self.norm2_name, norm2 = build_norm_layer(norm_cfg, planes, postfix=2)

        self.conv1 = build_conv_layer(
            conv_cfg,
            inplanes,
            planes,
            3,
            stride=stride,
            padding=dilation,
            dilation=dilation,
            bias=False)
        self.add_module(self.norm1_name, norm1)
        self.conv2 = build_conv_layer(
            conv_cfg, planes, planes, 3, padding=1, bias=False)
        self.add_module(self.norm2_name, norm2)

        self.relu = nn.ReLU(inplace=True)
        self.stride = stride
        self.dilation = dilation

    @property
    def norm1(self):
        return getattr(self, self.norm1_name)

    @property
    def norm2(self):
        return getattr(self, self.norm2_name)

    def forward(self, x, H, W):
        B, _, C = x.shape
        x = x.permute(0, 2, 1).reshape(B, -1, H, W)
        identity = x

        out = self.conv1(x)
        out = self.norm1(out)
        out = self.relu(out)

        out = self.conv2(out)
        out = self.norm2(out)

        out += identity
        out = self.relu(out)
        out = out.flatten(2).transpose(1, 2)
        return out


class Bottleneck(nn.Module):
    expansion = 4

    def __init__(self,
                 inplanes,
                 planes,
                 stride=1,
                 dilation=1,
                 conv_cfg=None,
                 norm_cfg=dict(type='BN')):
        """Bottleneck block for ResNet.
        If style is "pytorch", the stride-two layer is the 3x3 conv layer,
        if it is "caffe", the stride-two layer is the first 1x1 conv layer.
        """
        super(Bottleneck, self).__init__()

        self.inplanes = inplanes
        self.planes = planes
        self.stride = stride
        self.dilation = dilation
        self.conv_cfg = conv_cfg
        self.norm_cfg = norm_cfg

        self.conv1_stride = 1
        self.conv2_stride = stride

        self.norm1_name, norm1 = build_norm_layer(norm_cfg, planes, postfix=1)
        self.norm2_name, norm2 = build_norm_layer(norm_cfg, planes, postfix=2)
        self.norm3_name, norm3 = build_norm_layer(
            norm_cfg, planes * self.expansion, postfix=3)

        self.conv1 = build_conv_layer(
            conv_cfg,
            inplanes,
            planes,
            kernel_size=1,
            stride=self.conv1_stride,
            bias=False)
        self.add_module(self.norm1_name, norm1)
        self.conv2 = build_conv_layer(
            conv_cfg,
            planes,
            planes,
            kernel_size=3,
            stride=self.conv2_stride,
            padding=dilation,
            dilation=dilation,
            bias=False)
        self.add_module(self.norm2_name, norm2)
        self.conv3 = build_conv_layer(
            conv_cfg,
            planes,
            planes * self.expansion,
            kernel_size=1,
            bias=False)
        self.add_module(self.norm3_name, norm3)

        self.relu = nn.ReLU(inplace=True)

    @property
    def norm1(self):
        return getattr(self, self.norm1_name)

    @property
    def norm2(self):
        return getattr(self, self.norm2_name)

    @property
    def norm3(self):
        return getattr(self, self.norm3_name)

    def forward(self, x, H, W):
        B, _, C = x.shape
        x = x.permute(0, 2, 1).reshape(B, -1, H, W)
        identity = x

        out = self.conv1(x)
        out = self.norm1(out)
        out = self.relu(out)

        out = self.conv2(out)
        out = self.norm2(out)
        out = self.relu(out)

        out = self.conv3(out)
        out = self.norm3(out)

        out += identity
        out = self.relu(out)
        out = out.flatten(2).transpose(1, 2)
        return out


class Attention(nn.Module):

    def __init__(self,
                 dim,
                 num_heads=8,
                 qkv_bias=False,
                 qk_scale=None,
                 attn_drop=0.,
                 proj_drop=0.,
                 window_size=None,
                 attn_head_dim=None):
        super().__init__()
        self.num_heads = num_heads
        head_dim = dim // num_heads
        if attn_head_dim is not None:
            head_dim = attn_head_dim
        all_head_dim = head_dim * self.num_heads
        # NOTE scale factor was wrong in my original version, can set manually to be compat with prev weights
        self.scale = qk_scale or head_dim**-0.5

        self.qkv = nn.Linear(dim, all_head_dim * 3, bias=qkv_bias)
        self.window_size = window_size
        q_size = window_size[0]
        kv_size = q_size
        rel_sp_dim = 2 * q_size - 1
        self.rel_pos_h = nn.Parameter(torch.zeros(rel_sp_dim, head_dim))
        self.rel_pos_w = nn.Parameter(torch.zeros(rel_sp_dim, head_dim))

        self.attn_drop = nn.Dropout(attn_drop)
        self.proj = nn.Linear(all_head_dim, dim)
        self.proj_drop = nn.Dropout(proj_drop)

    def forward(self, x, H, W, rel_pos_bias=None):
        B, N, C = x.shape
        # qkv_bias = None
        # if self.q_bias is not None:
        #     qkv_bias = torch.cat((self.q_bias, torch.zeros_like(self.v_bias, requires_grad=False), self.v_bias))
        # qkv = self.qkv(x).reshape(B, N, 3, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4)
        qkv = self.qkv(x)
        qkv = qkv.reshape(B, N, 3, self.num_heads, -1).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))
        attn = calc_rel_pos_spatial(attn, q, self.window_size,
                                    self.window_size, self.rel_pos_h,
                                    self.rel_pos_w)
        # if self.relative_position_bias_table is not None:
        #     relative_position_bias = \
        #         self.relative_position_bias_table[self.relative_position_index.view(-1)].view(
        #             self.window_size[0] * self.window_size[1] + 1,
        #             self.window_size[0] * self.window_size[1] + 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 rel_pos_bias is not None:
        #     attn = attn + rel_pos_bias

        attn = attn.softmax(dim=-1)
        attn = self.attn_drop(attn)

        x = (attn @ v).transpose(1, 2).reshape(B, N, -1)
        x = self.proj(x)
        x = self.proj_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


def calc_rel_pos_spatial(
    attn,
    q,
    q_shape,
    k_shape,
    rel_pos_h,
    rel_pos_w,
):
    """
    Spatial Relative Positional Embeddings.
    """
    sp_idx = 0
    q_h, q_w = q_shape
    k_h, k_w = k_shape

    # Scale up rel pos if shapes for q and k are different.
    q_h_ratio = max(k_h / q_h, 1.0)
    k_h_ratio = max(q_h / k_h, 1.0)
    dist_h = (
        torch.arange(q_h)[:, None] * q_h_ratio -
        torch.arange(k_h)[None, :] * k_h_ratio)
    dist_h += (k_h - 1) * k_h_ratio
    q_w_ratio = max(k_w / q_w, 1.0)
    k_w_ratio = max(q_w / k_w, 1.0)
    dist_w = (
        torch.arange(q_w)[:, None] * q_w_ratio -
        torch.arange(k_w)[None, :] * k_w_ratio)
    dist_w += (k_w - 1) * k_w_ratio

    Rh = rel_pos_h[dist_h.long()]
    Rw = rel_pos_w[dist_w.long()]

    B, n_head, q_N, dim = q.shape

    r_q = q[:, :, sp_idx:].reshape(B, n_head, q_h, q_w, dim)
    rel_h = torch.einsum('byhwc,hkc->byhwk', r_q, Rh)
    rel_w = torch.einsum('byhwc,wkc->byhwk', r_q, Rw)

    attn[:, :, sp_idx:, sp_idx:] = (
        attn[:, :, sp_idx:, sp_idx:].view(B, -1, q_h, q_w, k_h, k_w) +
        rel_h[:, :, :, :, :, None] + rel_w[:, :, :, :, None, :]).view(
            B, -1, q_h * q_w, k_h * k_w)

    return attn


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.,
                 attn_head_dim=None):

        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

        q_size = window_size[0]
        kv_size = window_size[1]
        rel_sp_dim = 2 * q_size - 1
        self.rel_pos_h = nn.Parameter(torch.zeros(rel_sp_dim, head_dim))
        self.rel_pos_w = nn.Parameter(torch.zeros(rel_sp_dim, head_dim))

        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, H, W):
        """ 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
        x = x.reshape(B_, H, W, C)
        pad_l = pad_t = 0
        pad_r = (self.window_size[1] -
                 W % self.window_size[1]) % self.window_size[1]
        pad_b = (self.window_size[0] -
                 H % self.window_size[0]) % self.window_size[0]

        x = F.pad(x, (0, 0, pad_l, pad_r, pad_t, pad_b))
        _, Hp, Wp, _ = x.shape

        x = window_partition(
            x, self.window_size[0])  # nW*B, window_size, window_size, C
        x = x.view(-1, self.window_size[1] * self.window_size[0],
                   C)  # nW*B, window_size*window_size, C
        B_w = x.shape[0]
        N_w = x.shape[1]
        qkv = self.qkv(x).reshape(B_w, N_w, 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))

        attn = calc_rel_pos_spatial(attn, q, self.window_size,
                                    self.window_size, self.rel_pos_h,
                                    self.rel_pos_w)

        attn = self.softmax(attn)

        attn = self.attn_drop(attn)

        x = (attn @ v).transpose(1, 2).reshape(B_w, N_w, C)
        x = self.proj(x)
        x = self.proj_drop(x)

        x = x.view(-1, self.window_size[1], self.window_size[0], C)
        x = window_reverse(x, self.window_size[0], Hp, Wp)  # B H' W' C

        if pad_r > 0 or pad_b > 0:
            x = x[:, :H, :W, :].contiguous()

        x = x.view(B_, H * W, C)

        return x


class Block(nn.Module):

    def __init__(self,
                 dim,
                 num_heads,
                 mlp_ratio=4.,
                 qkv_bias=False,
                 qk_scale=None,
                 drop=0.,
                 attn_drop=0.,
                 drop_path=0.,
                 init_values=None,
                 act_layer=nn.GELU,
                 norm_layer=nn.LayerNorm,
                 window_size=None,
                 attn_head_dim=None,
                 window=False,
                 aggregation='attn'):
        super().__init__()
        self.norm1 = norm_layer(dim)
        self.aggregation = aggregation
        self.window = window
        if not window:
            if aggregation == 'attn':
                self.attn = Attention(
                    dim,
                    num_heads=num_heads,
                    qkv_bias=qkv_bias,
                    qk_scale=qk_scale,
                    attn_drop=attn_drop,
                    proj_drop=drop,
                    window_size=window_size,
                    attn_head_dim=attn_head_dim)
            else:
                self.attn = WindowAttention(
                    dim,
                    num_heads=num_heads,
                    qkv_bias=qkv_bias,
                    qk_scale=qk_scale,
                    attn_drop=attn_drop,
                    proj_drop=drop,
                    window_size=window_size,
                    attn_head_dim=attn_head_dim)
                if aggregation == 'basicblock':
                    self.conv_aggregation = BasicBlock(
                        inplanes=dim, planes=dim)
                elif aggregation == 'bottleneck':
                    self.conv_aggregation = Bottleneck(
                        inplanes=dim, planes=dim // 4)
        else:
            self.attn = WindowAttention(
                dim,
                num_heads=num_heads,
                qkv_bias=qkv_bias,
                qk_scale=qk_scale,
                attn_drop=attn_drop,
                proj_drop=drop,
                window_size=window_size,
                attn_head_dim=attn_head_dim)
        # NOTE: drop path for stochastic depth, we shall see if this is better than dropout here
        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)

        if init_values is not None:
            self.gamma_1 = nn.Parameter(
                init_values * torch.ones((dim)), requires_grad=True)
            self.gamma_2 = nn.Parameter(
                init_values * torch.ones((dim)), requires_grad=True)
        else:
            self.gamma_1, self.gamma_2 = None, None

    def forward(self, x, H, W):
        if self.gamma_1 is None:
            x = x + self.drop_path(self.attn(self.norm1(x), H, W))
            x = x + self.drop_path(self.mlp(self.norm2(x)))
        else:
            x = x + self.drop_path(
                self.gamma_1 * self.attn(self.norm1(x), H, W))
            x = x + self.drop_path(self.gamma_2 * self.mlp(self.norm2(x)))
        if not self.window and self.aggregation != 'attn':
            x = self.conv_aggregation(x, H, W)
        return x


class PatchEmbed(nn.Module):
    """ Image to Patch Embedding
    """

    def __init__(self, img_size=224, patch_size=16, in_chans=3, embed_dim=768):
        super().__init__()
        img_size = to_2tuple(img_size)
        patch_size = to_2tuple(patch_size)
        num_patches = (img_size[1] // patch_size[1]) * (
            img_size[0] // patch_size[0])
        self.patch_shape = (img_size[0] // patch_size[0],
                            img_size[1] // patch_size[1])
        self.img_size = img_size
        self.patch_size = patch_size
        self.num_patches = num_patches

        self.proj = nn.Conv2d(
            in_chans, embed_dim, kernel_size=patch_size, stride=patch_size)

    def forward(self, x, **kwargs):
        B, C, H, W = x.shape
        # FIXME look at relaxing size constraints
        # assert H == self.img_size[0] and W == self.img_size[1], \
        #     f"Input image size ({H}*{W}) doesn't match model ({self.img_size[0]}*{self.img_size[1]})."
        x = self.proj(x)
        Hp, Wp = x.shape[2], x.shape[3]

        x = x.flatten(2).transpose(1, 2)
        return x, (Hp, Wp)


class HybridEmbed(nn.Module):
    """ CNN Feature Map Embedding
    Extract feature map from CNN, flatten, project to embedding dim.
    """

    def __init__(self,
                 backbone,
                 img_size=224,
                 feature_size=None,
                 in_chans=3,
                 embed_dim=768):
        super().__init__()
        assert isinstance(backbone, nn.Module)
        img_size = to_2tuple(img_size)
        self.img_size = img_size
        self.backbone = backbone
        if feature_size is None:
            with torch.no_grad():
                # FIXME this is hacky, but most reliable way of determining the exact dim of the output feature
                # map for all networks, the feature metadata has reliable channel and stride info, but using
                # stride to calc feature dim requires info about padding of each stage that isn't captured.
                training = backbone.training
                if training:
                    backbone.eval()
                o = self.backbone(
                    torch.zeros(1, in_chans, img_size[0], img_size[1]))[-1]
                feature_size = o.shape[-2:]
                feature_dim = o.shape[1]
                backbone.train(training)
        else:
            feature_size = to_2tuple(feature_size)
            feature_dim = self.backbone.feature_info.channels()[-1]
        self.num_patches = feature_size[0] * feature_size[1]
        self.proj = nn.Linear(feature_dim, embed_dim)

    def forward(self, x):
        x = self.backbone(x)[-1]
        x = x.flatten(2).transpose(1, 2)
        x = self.proj(x)
        return x


class Norm2d(nn.Module):

    def __init__(self, embed_dim):
        super().__init__()
        self.ln = nn.LayerNorm(embed_dim, eps=1e-6)

    def forward(self, x):
        x = x.permute(0, 2, 3, 1)
        x = self.ln(x)
        x = x.permute(0, 3, 1, 2).contiguous()
        return x


# todo: refactor vitdet and vit_transformer_dynamic
@BACKBONES.register_module()
class ViTDet(nn.Module):
    """ Vision Transformer with support for patch or hybrid CNN input stage
    """

    def __init__(self,
                 img_size=224,
                 patch_size=16,
                 in_chans=3,
                 num_classes=80,
                 embed_dim=768,
                 depth=12,
                 num_heads=12,
                 mlp_ratio=4.,
                 qkv_bias=False,
                 qk_scale=None,
                 drop_rate=0.,
                 attn_drop_rate=0.,
                 drop_path_rate=0.,
                 hybrid_backbone=None,
                 norm_layer=None,
                 init_values=None,
                 use_checkpoint=False,
                 use_abs_pos_emb=False,
                 use_rel_pos_bias=False,
                 use_shared_rel_pos_bias=False,
                 out_indices=[11],
                 interval=3,
                 pretrained=None,
                 aggregation='attn'):
        super().__init__()
        norm_layer = norm_layer or partial(nn.LayerNorm, eps=1e-6)
        self.num_classes = num_classes
        self.num_features = self.embed_dim = embed_dim  # num_features for consistency with other models

        if hybrid_backbone is not None:
            self.patch_embed = HybridEmbed(
                hybrid_backbone,
                img_size=img_size,
                in_chans=in_chans,
                embed_dim=embed_dim)
        else:
            self.patch_embed = PatchEmbed(
                img_size=img_size,
                patch_size=patch_size,
                in_chans=in_chans,
                embed_dim=embed_dim)

        num_patches = self.patch_embed.num_patches

        self.out_indices = out_indices

        if use_abs_pos_emb:
            self.pos_embed = nn.Parameter(
                torch.zeros(1, num_patches, embed_dim))
        else:
            self.pos_embed = None

        self.pos_drop = nn.Dropout(p=drop_rate)

        dpr = [x.item() for x in torch.linspace(0, drop_path_rate, depth)
               ]  # stochastic depth decay rule
        self.use_rel_pos_bias = use_rel_pos_bias
        self.use_checkpoint = use_checkpoint
        self.blocks = nn.ModuleList([
            Block(
                dim=embed_dim,
                num_heads=num_heads,
                mlp_ratio=mlp_ratio,
                qkv_bias=qkv_bias,
                qk_scale=qk_scale,
                drop=drop_rate,
                attn_drop=attn_drop_rate,
                drop_path=dpr[i],
                norm_layer=norm_layer,
                init_values=init_values,
                window_size=(14, 14) if
                ((i + 1) % interval != 0
                 or aggregation != 'attn') else self.patch_embed.patch_shape,
                window=((i + 1) % interval != 0),
                aggregation=aggregation) for i in range(depth)
        ])

        if self.pos_embed is not None:
            trunc_normal_(self.pos_embed, std=.02)

        self.norm = norm_layer(embed_dim)

        self.pretrained = pretrained
        self._register_load_state_dict_pre_hook(self._prepare_checkpoint_hook)

    def fix_init_weight(self):

        def rescale(param, layer_id):
            param.div_(math.sqrt(2.0 * layer_id))

        for layer_id, layer in enumerate(self.blocks):
            rescale(layer.attn.proj.weight.data, layer_id + 1)
            rescale(layer.mlp.fc2.weight.data, layer_id + 1)

    def init_weights(self, pretrained=None):
        """Initialize the weights in backbone.
        Args:
            pretrained (str, optional): Path to pre-trained weights.
                Defaults to None.
        """
        self.fix_init_weight()
        pretrained = pretrained or self.pretrained

        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(m, nn.Conv2d):
                kaiming_init(m, mode='fan_in', nonlinearity='relu')
            elif isinstance(m, (_BatchNorm, nn.GroupNorm)):
                constant_init(m, 1)

            if isinstance(m, Bottleneck):
                constant_init(m.norm3, 0)
            elif isinstance(m, BasicBlock):
                constant_init(m.norm2, 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 _prepare_checkpoint_hook(self, state_dict, prefix, *args, **kwargs):
        rank, _ = get_dist_info()
        if 'pos_embed' in state_dict:
            pos_embed_checkpoint = state_dict['pos_embed']
            embedding_size = pos_embed_checkpoint.shape[-1]
            H, W = self.patch_embed.patch_shape
            num_patches = self.patch_embed.num_patches
            num_extra_tokens = 1
            # height (== width) for the checkpoint position embedding
            orig_size = int(
                (pos_embed_checkpoint.shape[-2] - num_extra_tokens)**0.5)
            # height (== width) for the new position embedding
            new_size = int(num_patches**0.5)
            # class_token and dist_token are kept unchanged
            if orig_size != new_size:
                if rank == 0:
                    print('Position interpolate from %dx%d to %dx%d' %
                          (orig_size, orig_size, H, W))
                # extra_tokens = pos_embed_checkpoint[:, :num_extra_tokens]
                # only the position tokens are interpolated
                pos_tokens = pos_embed_checkpoint[:, num_extra_tokens:]
                pos_tokens = pos_tokens.reshape(-1, orig_size, orig_size,
                                                embedding_size).permute(
                                                    0, 3, 1, 2)
                pos_tokens = torch.nn.functional.interpolate(
                    pos_tokens,
                    size=(H, W),
                    mode='bicubic',
                    align_corners=False)
                new_pos_embed = pos_tokens.permute(0, 2, 3, 1).flatten(1, 2)
                # new_pos_embed = torch.cat((extra_tokens, pos_tokens), dim=1)
                state_dict['pos_embed'] = new_pos_embed

    def get_num_layers(self):
        return len(self.blocks)

    @torch.jit.ignore
    def no_weight_decay(self):
        return {'pos_embed', 'cls_token'}

    def forward_features(self, x):
        B, C, H, W = x.shape
        x, (Hp, Wp) = self.patch_embed(x)
        batch_size, seq_len, _ = x.size()

        if self.pos_embed is not None:
            x = x + self.pos_embed
        x = self.pos_drop(x)

        outs = []
        for i, blk in enumerate(self.blocks):
            if self.use_checkpoint:
                x = checkpoint.checkpoint(blk, x)
            else:
                x = blk(x, Hp, Wp)

        x = self.norm(x)
        xp = x.permute(0, 2, 1).reshape(B, -1, Hp, Wp)

        outs.append(xp)

        return tuple(outs)

    def forward(self, x):
        x = self.forward_features(x)
        return x