models/video_swin_transformer.py

"""
Modified from Video-Swin-Transformer https://github.com/SwinTransformer/Video-Swin-Transformer
"""

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, trunc_normal_
from functools import reduce, lru_cache
from operator import mul
from einops import rearrange
from typing import Dict, List

from util.misc import NestedTensor
from .position_encoding import build_position_encoding

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, D, H, W, C)
        window_size (tuple[int]): window size

    Returns:
        windows: (B*num_windows, window_size*window_size, C)
    """
    B, D, H, W, C = x.shape
    x = x.view(B, D // window_size[0], window_size[0], H // window_size[1], window_size[1], W // window_size[2], window_size[2], C)
    windows = x.permute(0, 1, 3, 5, 2, 4, 6, 7).contiguous().view(-1, reduce(mul, window_size), C)
    return windows


def window_reverse(windows, window_size, B, D, H, W):
    """
    Args:
        windows: (B*num_windows, window_size, window_size, C)
        window_size (tuple[int]): Window size
        H (int): Height of image
        W (int): Width of image

    Returns:
        x: (B, D, H, W, C)
    """
    x = windows.view(B, D // window_size[0], H // window_size[1], W // window_size[2], window_size[0], window_size[1], window_size[2], -1)
    x = x.permute(0, 1, 4, 2, 5, 3, 6, 7).contiguous().view(B, D, H, W, -1)
    return x


def get_window_size(x_size, window_size, shift_size=None):
    use_window_size = list(window_size)
    if shift_size is not None:
        use_shift_size = list(shift_size)
    for i in range(len(x_size)):
        if x_size[i] <= window_size[i]:
            use_window_size[i] = x_size[i]
            if shift_size is not None:
                use_shift_size[i] = 0

    if shift_size is None:
        return tuple(use_window_size)
    else:
        return tuple(use_window_size), tuple(use_shift_size)


class WindowAttention3D(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 temporal length, 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=False, qk_scale=None, attn_drop=0., proj_drop=0.):

        super().__init__()
        self.dim = dim
        self.window_size = window_size  # Wd, 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) * (2 * window_size[2] - 1), num_heads))  # 2*Wd-1 * 2*Wh-1 * 2*Ww-1, nH

        # get pair-wise relative position index for each token inside the window
        coords_d = torch.arange(self.window_size[0])
        coords_h = torch.arange(self.window_size[1])
        coords_w = torch.arange(self.window_size[2])
        coords = torch.stack(torch.meshgrid(coords_d, coords_h, coords_w))  # 3, Wd, Wh, Ww
        coords_flatten = torch.flatten(coords, 1)  # 3, Wd*Wh*Ww
        relative_coords = coords_flatten[:, :, None] - coords_flatten[:, None, :]  # 3, Wd*Wh*Ww, Wd*Wh*Ww
        relative_coords = relative_coords.permute(1, 2, 0).contiguous()  # Wd*Wh*Ww, Wd*Wh*Ww, 3
        relative_coords[:, :, 0] += self.window_size[0] - 1  # shift to start from 0
        relative_coords[:, :, 1] += self.window_size[1] - 1
        relative_coords[:, :, 2] += self.window_size[2] - 1

        relative_coords[:, :, 0] *= (2 * self.window_size[1] - 1) * (2 * self.window_size[2] - 1)
        relative_coords[:, :, 1] *= (2 * self.window_size[2] - 1)
        relative_position_index = relative_coords.sum(-1)  # Wd*Wh*Ww, Wd*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, N, N) 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]  # B_, nH, N, C

        q = q * self.scale
        attn = q @ k.transpose(-2, -1)

        relative_position_bias = self.relative_position_bias_table[self.relative_position_index[:N, :N].reshape(-1)].reshape(
            N, N, -1)  # Wd*Wh*Ww,Wd*Wh*Ww,nH
        relative_position_bias = relative_position_bias.permute(2, 0, 1).contiguous()  # nH, Wd*Wh*Ww, Wd*Wh*Ww
        attn = attn + relative_position_bias.unsqueeze(0) # B_, nH, N, N

        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 SwinTransformerBlock3D(nn.Module):
    """ Swin Transformer Block.

    Args:
        dim (int): Number of input channels.
        num_heads (int): Number of attention heads.
        window_size (tuple[int]): Window size.
        shift_size (tuple[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=(2,7,7), shift_size=(0,0,0),
                 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, use_checkpoint=False):
        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
        self.use_checkpoint=use_checkpoint

        assert 0 <= self.shift_size[0] < self.window_size[0], "shift_size must in 0-window_size"
        assert 0 <= self.shift_size[1] < self.window_size[1], "shift_size must in 0-window_size"
        assert 0 <= self.shift_size[2] < self.window_size[2], "shift_size must in 0-window_size"

        self.norm1 = norm_layer(dim)
        self.attn = WindowAttention3D(
            dim, window_size=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)

    def forward_part1(self, x, mask_matrix):
        B, D, H, W, C = x.shape
        window_size, shift_size = get_window_size((D, H, W), self.window_size, self.shift_size)

        x = self.norm1(x)
        # pad feature maps to multiples of window size
        pad_l = pad_t = pad_d0 = 0
        pad_d1 = (window_size[0] - D % window_size[0]) % window_size[0]
        pad_b = (window_size[1] - H % window_size[1]) % window_size[1]
        pad_r = (window_size[2] - W % window_size[2]) % window_size[2]
        x = F.pad(x, (0, 0, pad_l, pad_r, pad_t, pad_b, pad_d0, pad_d1))
        _, Dp, Hp, Wp, _ = x.shape
        # cyclic shift
        if any(i > 0 for i in shift_size):
            shifted_x = torch.roll(x, shifts=(-shift_size[0], -shift_size[1], -shift_size[2]), dims=(1, 2, 3))
            attn_mask = mask_matrix
        else:
            shifted_x = x
            attn_mask = None
        # partition windows
        x_windows = window_partition(shifted_x, window_size)  # B*nW, Wd*Wh*Ww, C
        # W-MSA/SW-MSA
        attn_windows = self.attn(x_windows, mask=attn_mask)  # B*nW, Wd*Wh*Ww, C
        # merge windows
        attn_windows = attn_windows.view(-1, *(window_size+(C,)))
        shifted_x = window_reverse(attn_windows, window_size, B, Dp, Hp, Wp)  # B D' H' W' C
        # reverse cyclic shift
        if any(i > 0 for i in shift_size):
            x = torch.roll(shifted_x, shifts=(shift_size[0], shift_size[1], shift_size[2]), dims=(1, 2, 3))
        else:
            x = shifted_x

        if pad_d1 >0 or pad_r > 0 or pad_b > 0:
            x = x[:, :D, :H, :W, :].contiguous()
        return x

    def forward_part2(self, x):
        return self.drop_path(self.mlp(self.norm2(x)))

    def forward(self, x, mask_matrix):
        """ Forward function.

        Args:
            x: Input feature, tensor size (B, D, H, W, C).
            mask_matrix: Attention mask for cyclic shift.
        """

        shortcut = x
        if self.use_checkpoint:
            x = checkpoint.checkpoint(self.forward_part1, x, mask_matrix)
        else:
            x = self.forward_part1(x, mask_matrix)
        x = shortcut + self.drop_path(x)

        if self.use_checkpoint:
            x = x + checkpoint.checkpoint(self.forward_part2, x)
        else:
            x = x + self.forward_part2(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):
        """ Forward function.

        Args:
            x: Input feature, tensor size (B, D, H, W, C).
        """
        B, D, H, W, C = x.shape

        # 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 D H/2 W/2 C
        x1 = x[:, :, 1::2, 0::2, :]  # B D H/2 W/2 C
        x2 = x[:, :, 0::2, 1::2, :]  # B D H/2 W/2 C
        x3 = x[:, :, 1::2, 1::2, :]  # B D H/2 W/2 C
        x = torch.cat([x0, x1, x2, x3], -1)  # B D H/2 W/2 4*C

        x = self.norm(x)
        x = self.reduction(x)

        return x


# cache each stage results
@lru_cache()
def compute_mask(D, H, W, window_size, shift_size, device):
    img_mask = torch.zeros((1, D, H, W, 1), device=device)  # 1 Dp Hp Wp 1
    cnt = 0
    for d in slice(-window_size[0]), slice(-window_size[0], -shift_size[0]), slice(-shift_size[0],None):
        for h in slice(-window_size[1]), slice(-window_size[1], -shift_size[1]), slice(-shift_size[1],None):
            for w in slice(-window_size[2]), slice(-window_size[2], -shift_size[2]), slice(-shift_size[2],None):
                img_mask[:, d, h, w, :] = cnt
                cnt += 1
    mask_windows = window_partition(img_mask, window_size)  # nW, ws[0]*ws[1]*ws[2], 1
    mask_windows = mask_windows.squeeze(-1)  # nW, ws[0]*ws[1]*ws[2]
    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))
    return attn_mask


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 (tuple[int]): Local window size. Default: (1,7,7).
        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
    """

    def __init__(self,
                 dim,
                 depth,
                 num_heads,
                 window_size=(1,7,7),
                 mlp_ratio=4.,
                 qkv_bias=False,
                 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 = tuple(i // 2 for i in window_size)
        self.depth = depth
        self.use_checkpoint = use_checkpoint

        # build blocks
        self.blocks = nn.ModuleList([
            SwinTransformerBlock3D(
                dim=dim,
                num_heads=num_heads,
                window_size=window_size,
                shift_size=(0,0,0) if (i % 2 == 0) else self.shift_size,
                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,
                use_checkpoint=use_checkpoint,
            )
            for i in range(depth)])
        
        self.downsample = downsample
        if self.downsample is not None:
            self.downsample = downsample(dim=dim, norm_layer=norm_layer)

    def forward(self, x):
        """ Forward function.

        Args:
            x: Input feature, tensor size (B, C, D, H, W).
        """
        # calculate attention mask for SW-MSA
        B, C, D, H, W = x.shape
        window_size, shift_size = get_window_size((D,H,W), self.window_size, self.shift_size)
        x = rearrange(x, 'b c d h w -> b d h w c')
        Dp = int(np.ceil(D / window_size[0])) * window_size[0]
        Hp = int(np.ceil(H / window_size[1])) * window_size[1]
        Wp = int(np.ceil(W / window_size[2])) * window_size[2]
        attn_mask = compute_mask(Dp, Hp, Wp, window_size, shift_size, x.device)
        for blk in self.blocks:
            x = blk(x, attn_mask)
        x = x.view(B, D, H, W, -1)

        if self.downsample is not None:
            x = self.downsample(x)
        x = rearrange(x, 'b d h w c -> b c d h w')
        return x


class PatchEmbed3D(nn.Module):
    """ Video to Patch Embedding.

    Args:
        patch_size (int): Patch token size. Default: (2,4,4).
        in_chans (int): Number of input video 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=(2,4,4), in_chans=3, embed_dim=96, norm_layer=None):
        super().__init__()
        self.patch_size = patch_size

        self.in_chans = in_chans
        self.embed_dim = embed_dim

        self.proj = nn.Conv3d(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
        _, _, D, H, W = x.size()
        if W % self.patch_size[2] != 0:
            x = F.pad(x, (0, self.patch_size[2] - W % self.patch_size[2]))
        if H % self.patch_size[1] != 0:
            x = F.pad(x, (0, 0, 0, self.patch_size[1] - H % self.patch_size[1]))
        if D % self.patch_size[0] != 0:
            x = F.pad(x, (0, 0, 0, 0, 0, self.patch_size[0] - D % self.patch_size[0]))

        x = self.proj(x)  # B C D Wh Ww
        if self.norm is not None:
            D, Wh, Ww = x.size(2), x.size(3), x.size(4)
            x = x.flatten(2).transpose(1, 2)
            x = self.norm(x)
            x = x.transpose(1, 2).view(-1, self.embed_dim, D, Wh, Ww)

        return x


class SwinTransformer3D(nn.Module):
    """ Swin Transformer backbone.
        A PyTorch impl of : `Swin Transformer: Hierarchical Vision Transformer using Shifted Windows`  -
          https://arxiv.org/pdf/2103.14030

    Args:
        patch_size (int | tuple(int)): Patch size. Default: (4,4,4).
        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: Truee
        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: Normalization layer. Default: nn.LayerNorm.
        patch_norm (bool): If True, add normalization after patch embedding. Default: False.
        frozen_stages (int): Stages to be frozen (stop grad and set eval mode).
            -1 means not freezing any parameters.
    """

    def __init__(self,
                 pretrained=None,
                 pretrained2d=True,
                 patch_size=(4,4,4),
                 in_chans=3,
                 embed_dim=96,
                 depths=[2, 2, 6, 2],
                 num_heads=[3, 6, 12, 24],
                 window_size=(2,7,7),
                 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,
                 patch_norm=False,
                 frozen_stages=-1,
                 use_checkpoint=False):
        super().__init__()

        self.pretrained = pretrained
        self.pretrained2d = pretrained2d
        self.num_layers = len(depths)
        self.embed_dim = embed_dim
        self.patch_norm = patch_norm
        self.frozen_stages = frozen_stages
        self.window_size = window_size
        self.patch_size = patch_size

        # split image into non-overlapping patches
        self.patch_embed = PatchEmbed3D(
            patch_size=patch_size, in_chans=in_chans, embed_dim=embed_dim,
            norm_layer=norm_layer if self.patch_norm else None)

        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)

        self.num_features = int(embed_dim * 2**(self.num_layers-1))

        # add a norm layer for each output
        self.norm = norm_layer(self.num_features)

        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:
            self.pos_drop.eval()
            for i in range(0, self.frozen_stages):
                m = self.layers[i]
                m.eval()
                for param in m.parameters():
                    param.requires_grad = False

    def inflate_weights(self, logger):
        """Inflate the swin2d parameters to swin3d.

        The differences between swin3d and swin2d mainly lie in an extra
        axis. To utilize the pretrained parameters in 2d model,
        the weight of swin2d models should be inflated to fit in the shapes of
        the 3d counterpart.

        Args:
            logger (logging.Logger): The logger used to print
                debugging infomation.
        """
        checkpoint = torch.load(self.pretrained, map_location='cpu')
        state_dict = checkpoint['model']

        # delete relative_position_index since we always re-init it
        relative_position_index_keys = [k for k in state_dict.keys() if "relative_position_index" in k]
        for k in relative_position_index_keys:
            del state_dict[k]

        # delete attn_mask since we always re-init it
        attn_mask_keys = [k for k in state_dict.keys() if "attn_mask" in k]
        for k in attn_mask_keys:
            del state_dict[k]

        state_dict['patch_embed.proj.weight'] = state_dict['patch_embed.proj.weight'].unsqueeze(2).repeat(1,1,self.patch_size[0],1,1) / self.patch_size[0]

        # bicubic interpolate relative_position_bias_table if not match
        relative_position_bias_table_keys = [k for k in state_dict.keys() if "relative_position_bias_table" in k]
        for k in relative_position_bias_table_keys:
            relative_position_bias_table_pretrained = state_dict[k]
            relative_position_bias_table_current = self.state_dict()[k]
            L1, nH1 = relative_position_bias_table_pretrained.size()
            L2, nH2 = relative_position_bias_table_current.size()
            L2 = (2*self.window_size[1]-1) * (2*self.window_size[2]-1)
            wd = self.window_size[0]
            if nH1 != nH2:
                logger.warning(f"Error in loading {k}, passing")
            else:
                if L1 != L2:
                    S1 = int(L1 ** 0.5)
                    relative_position_bias_table_pretrained_resized = torch.nn.functional.interpolate(
                        relative_position_bias_table_pretrained.permute(1, 0).view(1, nH1, S1, S1), size=(2*self.window_size[1]-1, 2*self.window_size[2]-1),
                        mode='bicubic')
                    relative_position_bias_table_pretrained = relative_position_bias_table_pretrained_resized.view(nH2, L2).permute(1, 0)
            state_dict[k] = relative_position_bias_table_pretrained.repeat(2*wd-1,1)

        msg = self.load_state_dict(state_dict, strict=False)
        logger.info(msg)
        logger.info(f"=> loaded successfully '{self.pretrained}'")
        del checkpoint
        torch.cuda.empty_cache()

    def forward(self, x):
        """Forward function."""
        x = self.patch_embed(x)

        x = self.pos_drop(x)

        for layer in self.layers:
            x = layer(x.contiguous())

        x = rearrange(x, 'n c d h w -> n d h w c')
        x = self.norm(x)
        x = rearrange(x, 'n d h w c -> n c d h w')

        return x  # [B, C, T, H, W], T does not downsample so we can get the per-frame feature

    def train(self, mode=True):
        """Convert the model into training mode while keep layers freezed."""
        super(SwinTransformer3D, self).train(mode)
        self._freeze_stages()



class VideoSwinTransformerBackbone(nn.Module):
    """
    A wrapper which allows using Video-Swin Transformer as a temporal encoder for MTTR.
    Check out video-swin's original paper at: https://arxiv.org/abs/2106.13230 for more info about this architecture.
    Only the 'tiny' version of video swin was tested and is currently supported in our project.
    Additionally, we slightly modify video-swin to make it output per-frame embeddings as required by MTTR (check our
    paper's supplementary for more details), and completely discard of its 4th block.
    """
    def __init__(self, backbone_pretrained: bool, backbone_pretrained_path, train_backbone: bool, **kwargs):
        super(VideoSwinTransformerBackbone, self).__init__()
        # patch_size is (1, 4, 4) instead of the original (2, 4, 4).
        # this prevents swinT's original temporal downsampling so we can get per-frame features.
        swin_backbone = SwinTransformer3D(**kwargs)
        if backbone_pretrained and isinstance(backbone_pretrained_path, str):
            state_dict = torch.load(backbone_pretrained_path)['state_dict']
            # extract swinT's kinetics-400 pretrained weights and ignore the rest (prediction head etc.)
            state_dict = {k[9:]: v for k, v in state_dict.items() if 'backbone.' in k}

            # sum over the patch embedding weight temporal dim  [96, 3, 2, 4, 4] --> [96, 3, 1, 4, 4]
            patch_embed_weight = state_dict['patch_embed.proj.weight']
            patch_embed_weight = patch_embed_weight.sum(dim=2, keepdims=True)
            state_dict['patch_embed.proj.weight'] = patch_embed_weight
            print(f'load from {backbone_pretrained_path}.') 
            swin_backbone.load_state_dict(state_dict)

        self.patch_embed = swin_backbone.patch_embed
        self.pos_drop = swin_backbone.pos_drop
        self.layers = swin_backbone.layers
        self.downsamples = nn.ModuleList()
        for layer in self.layers:
            self.downsamples.append(layer.downsample)
            layer.downsample = None
        self.downsamples[-1] = None  # downsampling after the last layer is not necessary

        self.layer_output_channels = [swin_backbone.embed_dim * 2 ** i for i in range(len(self.layers))]
        self.train_backbone = train_backbone
        if not train_backbone:
            for parameter in self.parameters():
                parameter.requires_grad_(False)

    def forward(self, samples: torch.Tensor, num_frames):
        # num_frames is needed, because we put time in batch dimension.
        # samples: [B*T, 3, H, W]
        n, c, h, w = samples.shape
        samples = rearrange(samples, '(b t) c h w -> b c t h w', b=n//num_frames, t=num_frames)
        vid_embeds = self.patch_embed(samples)
        vid_embeds = self.pos_drop(vid_embeds)

        out = {} # layer outputs before downsampling
        for idx, (layer, downsample) in enumerate(zip(self.layers, self.downsamples)):
            vid_embeds = layer(vid_embeds.contiguous())
            out[str(idx)] = vid_embeds
            if downsample:
                vid_embeds = rearrange(vid_embeds, 'b c t h w -> b t h w c')
                vid_embeds = downsample(vid_embeds)
                vid_embeds = rearrange(vid_embeds, 'b t h w c -> b c t h w')
        # rearrange
        for idx, o in out.items():
            out[idx] = rearrange(o, 'b c t h w -> (b t) c h w')
        return out



class BackboneBase(nn.Module):
    def __init__(self, backbone: nn.Module, strides=[4, 8, 16, 32], num_channels=[96, 192, 384, 768]):
        super().__init__()
        self.strides = strides
        self.num_channels = num_channels
        self.body = backbone

    def forward(self, tensor_list: NestedTensor, num_frames: int):
        xs = self.body(tensor_list.tensors, num_frames)
        out: Dict[str, NestedTensor] = {}
        for name, x in xs.items():
            m = tensor_list.mask
            assert m is not None
            mask = F.interpolate(m[None].float(), size=x.shape[-2:]).to(torch.bool)[0]
            out[name] = NestedTensor(x, mask)
        return out

class Backbone(BackboneBase):
    """ResNet backbone with frozen BatchNorm."""
    def __init__(self, name: str,
                 checkpoint: bool = False,
                 pretrained: str = None):
        assert name in ['video_swin_t_p4w7', 'video_swin_s_p4w7', 'video_swin_b_p4w7']
        cfgs = configs[name]
        cfgs.update({'use_checkpoint': checkpoint})
        out_indices = (0, 1, 2, 3)
        strides = [int(2**(i+2)) for i in out_indices]
        num_channels = [int(cfgs['embed_dim'] * 2**i) for i in out_indices]
        backbone = VideoSwinTransformerBackbone(True, pretrained, True, **cfgs)
        super().__init__(backbone, strides, num_channels)


configs = {
    'video_swin_t_p4w7': 
                   dict(patch_size=(1,4,4),
                        embed_dim=96,
                        depths=[2, 2, 6, 2],
                        num_heads=[3, 6, 12, 24],
                        window_size=(8,7,7),
                        mlp_ratio=4.,
                        qkv_bias=True,
                        qk_scale=None,
                        drop_rate=0.,
                        attn_drop_rate=0.,
                        drop_path_rate=0.2,
                        patch_norm=True,
                        use_checkpoint=False
                        ),
    'video_swin_s_p4w7': 
                   dict(patch_size=(1,4,4),
                        embed_dim=96,
                        depths=[2, 2, 18, 2],
                        num_heads=[3, 6, 12, 24],
                        window_size=(8,7,7),
                        mlp_ratio=4.,
                        qkv_bias=True,
                        qk_scale=None,
                        drop_rate=0.,
                        attn_drop_rate=0.,
                        drop_path_rate=0.2,
                        patch_norm=True,
                        use_checkpoint=False
                        ),
    'video_swin_b_p4w7': 
                   dict(patch_size=(1,4,4),
                        embed_dim=128,
                        depths=[2, 2, 18, 2],
                        num_heads=[4, 8, 16, 32],
                        window_size=(8,7,7),
                        mlp_ratio=4.,
                        qkv_bias=True,
                        qk_scale=None,
                        drop_rate=0.,
                        attn_drop_rate=0.,
                        drop_path_rate=0.2,
                        patch_norm=True,
                        use_checkpoint=False
                        )
}

class Joiner(nn.Sequential):
    def __init__(self, backbone, position_embedding):
        super().__init__(backbone, position_embedding)
        self.strides = backbone.strides
        self.num_channels = backbone.num_channels

    def forward(self, tensor_list: NestedTensor):
        _, t = tensor_list.tensors.shape[:2]
        tensor_list.tensors = rearrange(tensor_list.tensors, 'b t c h w -> (b t) c h w')
        tensor_list.mask = rearrange(tensor_list.mask, 'b t h w -> (b t) h w')

        xs = self[0](tensor_list, num_frames=t)
        out: List[NestedTensor] = []
        pos = []
        for name, x in sorted(xs.items()):
            out.append(x)
        # position encoding
        for x in out:
            pos.append(self[1](x).to(x.tensors.dtype))
        return out, pos

    
def build_video_swin_backbone(args):
    position_embedding = build_position_encoding(args)
    backbone = Backbone(args.backbone, args.use_checkpoint, args.backbone_pretrained)
    model = Joiner(backbone, position_embedding)
    return model
    

if __name__ == '__main__':
    cfgs = configs['video_swin_t_p4w7']
    model = VideoSwinTransformerBackbone(True, 'video_swin_pretrained/swin_tiny_patch244_window877_kinetics400_1k.pth', True, **cfgs).cuda()
    inputs = torch.randn(10, 3,384,224).cuda() # 10 = 2 x 5
    import ipdb; ipdb.set_trace()
    # outs
    # 0: (10, 96, 96, 56)
    # 1: (10, 192, 48, 28)
    # 2: (10, 384, 24, 14)
    # 3: (10, 768, 12, 7)
    out = model(inputs, num_frames=5)