Added architecture definitions

archs/swin3d_rcab.py

@@ -0,0 +1,881 @@
+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 mmcv.runner import load_checkpoint
+# from mmaction.utils import get_root_logger
+# from ..builder import BACKBONES
+
+from functools import reduce, lru_cache
+from operator import mul
+from einops import rearrange
+import sys
+
+class Upsample(nn.Sequential):
+    """Upsample module.
+
+    Args:
+        scale (int): Scale factor. Supported scales: 2^n and 3.
+        num_feat (int): Channel number of intermediate features.
+    """
+
+    def __init__(self, scale, num_feat):
+        m = []
+        if (scale & (scale - 1)) == 0:  # scale = 2^n
+            for _ in range(int(math.log(scale, 2))):
+                m.append(nn.Conv2d(num_feat, 4 * num_feat, 3, 1, 1))
+                m.append(nn.PixelShuffle(2))
+        elif scale == 3:
+            m.append(nn.Conv2d(num_feat, 9 * num_feat, 3, 1, 1))
+            m.append(nn.PixelShuffle(3))
+        else:
+            raise ValueError(f'scale {scale} is not supported. ' 'Supported scales: 2^n and 3.')
+        super(Upsample, self).__init__(*m)
+
+
+
+def make_layer(basic_block, num_basic_block, **kwarg):
+    """Make layers by stacking the same blocks.
+
+    Args:
+        basic_block (nn.module): nn.module class for basic block.
+        num_basic_block (int): number of blocks.
+
+    Returns:
+        nn.Sequential: Stacked blocks in nn.Sequential.
+    """
+    layers = []
+    for _ in range(num_basic_block):
+        layers.append(basic_block(**kwarg))
+    return nn.Sequential(*layers)
+
+
+class ChannelAttention(nn.Module):
+    """Channel attention used in RCAN.
+
+    Args:
+        num_feat (int): Channel number of intermediate features.
+        squeeze_factor (int): Channel squeeze factor. Default: 16.
+    """
+
+    def __init__(self, num_feat, squeeze_factor=16):
+        super(ChannelAttention, self).__init__()
+        self.attention = nn.Sequential(
+            nn.AdaptiveAvgPool2d(1), nn.Conv2d(num_feat, num_feat // squeeze_factor, 1, padding=0),
+            nn.ReLU(inplace=True), nn.Conv2d(num_feat // squeeze_factor, num_feat, 1, padding=0), nn.Sigmoid())
+
+    def forward(self, x):
+        y = self.attention(x)
+        return x * y
+
+
+class RCAB(nn.Module):
+    """Residual Channel Attention Block (RCAB) used in RCAN.
+
+    Args:
+        num_feat (int): Channel number of intermediate features.
+        squeeze_factor (int): Channel squeeze factor. Default: 16.
+        res_scale (float): Scale the residual. Default: 1.
+    """
+
+    def __init__(self, num_feat, squeeze_factor=16, res_scale=1):
+        super(RCAB, self).__init__()
+        self.res_scale = res_scale
+
+        self.rcab = nn.Sequential(
+            nn.Conv2d(num_feat, num_feat, 3, 1, 1), nn.ReLU(True), nn.Conv2d(num_feat, num_feat, 3, 1, 1),
+            ChannelAttention(num_feat, squeeze_factor))
+
+    def forward(self, x):
+        res = self.rcab(x) * self.res_scale
+        return res + x
+
+
+class ResidualGroup(nn.Module):
+    """Residual Group of RCAB.
+
+    Args:
+        num_feat (int): Channel number of intermediate features.
+        num_block (int): Block number in the body network.
+        squeeze_factor (int): Channel squeeze factor. Default: 16.
+        res_scale (float): Scale the residual. Default: 1.
+    """
+
+    def __init__(self, num_feat, num_block, squeeze_factor=16, res_scale=1):
+        super(ResidualGroup, self).__init__()
+
+        self.residual_group = make_layer(
+            RCAB, num_block, num_feat=num_feat, squeeze_factor=squeeze_factor, res_scale=res_scale)
+        self.conv = nn.Conv2d(num_feat, num_feat, 3, 1, 1)
+
+    def forward(self, x):
+        res = self.conv(self.residual_group(x))
+        return res + x
+
+
+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 RSTB3D(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,
+                 in_chans=1,
+                 patch_norm=True,
+                 patch_size=(3,4,4),
+                 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
+
+        self.basic_layer = BasicLayer(
+                dim=dim,
+                depth=depth,
+                num_heads=num_heads,
+                window_size=window_size,
+                mlp_ratio=mlp_ratio,
+                qkv_bias=qkv_bias,
+                qk_scale=qk_scale,
+                drop=drop,
+                attn_drop=attn_drop,
+                drop_path=drop_path,
+                norm_layer=norm_layer,
+                # downsample=PatchMerging if i_layer<self.num_layers-1 else None,
+                downsample=None,
+                use_checkpoint=use_checkpoint)
+
+        self.resi_connection1 = nn.Conv2d(3, 64, 3, 1, 1)
+        self.resi_connection2 = ResidualGroup(num_feat=64,squeeze_factor=16,num_block=12)
+        self.resi_connection3 = nn.Conv2d(64, 3, 3, 1, 1)
+
+        # split image into non-overlapping patches
+        self.patch_embed = PatchEmbed3D(
+            patch_size=patch_size, in_chans=in_chans, embed_dim=dim,
+            norm_layer=norm_layer if patch_norm else None)
+
+        # split image into non-overlapping patches
+        self.patch_unembed = PatchUnEmbed3D(
+            patch_size=patch_size, in_chans=in_chans, embed_dim=dim,
+            norm_layer=norm_layer if patch_norm else None)
+
+        self.lrelu = nn.LeakyReLU(negative_slope=0.2, inplace=True)
+
+    def forward(self, x):
+        shortcut = x
+        x = self.basic_layer(x)
+        x = self.patch_unembed(x)
+        x = self.resi_connection1(x)
+        x = self.lrelu(x)
+        x = self.resi_connection2(x)
+        x = self.lrelu(x)
+        x = self.resi_connection3(x)
+        x = self.lrelu(x)
+        x = self.patch_embed(x)
+        x = x + shortcut
+        x = self.lrelu(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 (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=(3,4,4), in_chans=3, embed_dim=96, norm_layer=None):
+        super().__init__()
+        self.patch_size = patch_size
+
+        #print('received 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."""
+        x = x.unsqueeze(1) # assuming gray scale video frames are encoded as channels, now separate
+
+        x = self.proj(x)  # B C D Wh Ww
+        if self.norm is not None:
+            #print('ionside here with self.norm')
+            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 PatchUnEmbed3D(nn.Module):
+    def __init__(self, patch_size=(3,4,4), in_chans=3, embed_dim=96, norm_layer=nn.LayerNorm):
+        super().__init__()
+        self.patch_size = patch_size
+
+        self.in_chans = in_chans
+        self.embed_dim = embed_dim
+
+        unembed_dim = 1
+        self.unembed_dim = unembed_dim
+
+        self.proj = nn.ConvTranspose3d(embed_dim, unembed_dim, kernel_size=patch_size, stride=patch_size)
+        self.conv = nn.Conv2d(3*unembed_dim, 3, 3, 1, 1)
+
+        if norm_layer is not None:
+            self.norm = norm_layer(unembed_dim)
+        else:
+            self.norm = None
+
+        self.lrelu = nn.LeakyReLU(negative_slope=0.2, inplace=True)
+
+    def forward(self, x):
+
+        D, Wh, Ww = x.size(2), x.size(3), x.size(4)
+        # x = x.view(-1,self.embed_dim*D,Wh,Ww)
+        x = self.proj(x)
+
+        # 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.unembed_dim, D, Wh, Ww)
+
+
+
+        x = self.lrelu(x)
+        x = x.view(-1,3*D,4*Wh,4*Ww)
+        # x = x.flatten(start_dim=1,end_dim=2)
+        # x = x.view(-1,9,4*Wh,4*Ww) # 18 128 128
+        x = self.lrelu(self.conv(x)) # 64 128 128
+
+        return x
+
+class Upsampler(nn.Module):
+    def __init__(self, patch_size=(3,4,4), in_chans=3, embed_dim=96, norm_layer=nn.LayerNorm):
+        super().__init__()
+        self.patch_size = patch_size
+
+        self.in_chans = in_chans
+        self.embed_dim = embed_dim
+
+        self.expand = nn.Conv2d(9, 20, 3, 1, 1)
+
+        self.shuffle = nn.PixelShuffle(2)
+        self.fusion = nn.Conv2d(20//4, 1, 3, 1, 1)
+
+        self.lrelu = nn.LeakyReLU(negative_slope=0.2, inplace=True)
+
+    def forward(self, x):
+
+        # x = x.view(-1,self.embed_dim*D,Wh,Ww)
+        x = self.lrelu(self.expand(x))
+        x = self.shuffle(x) # 16 256 256
+        x = self.lrelu(self.fusion(x))
+
+        return x
+
+
+class SwinTransformer3D_RCAB(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,
+                 opt,
+                 patch_size=(4,4,4),
+                 in_chans=1,
+                 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=True,
+                 upscale=2,
+                 frozen_stages=-1,
+                 use_checkpoint=False,
+                 vis=False,
+                 **kwargs):
+        super().__init__()
+
+        self.num_layers = len(depths)
+        self.embed_dim = embed_dim
+        self.patch_norm = patch_norm
+        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)
+
+        # split image into non-overlapping patches
+        self.patch_unembed = PatchUnEmbed3D(
+            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 = RSTB3D(
+                dim=embed_dim,
+                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,
+                downsample=None,
+                in_chans=in_chans,
+                patch_size=patch_size,
+                patch_norm=patch_norm,
+                use_checkpoint=use_checkpoint)
+            self.layers.append(layer)
+
+        self.num_features = int(embed_dim)
+
+        # add a norm layer for each output
+        self.norm = norm_layer(self.num_features)
+
+        self.upsampler = Upsampler(embed_dim=embed_dim)
+        self.task = opt.task
+
+        self.segmentation_decode = nn.Conv2d(3, 4, 1)
+
+
+    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 pretrained:
+            self.pretrained = pretrained
+        # if isinstance(self.pretrained, str):
+            # self.apply(_init_weights)
+            # logger = get_root_logger()
+            # logger.info(f'load model from: {self.pretrained}')
+
+            # if self.pretrained2d:
+                # Inflate 2D model into 3D model.
+                # self.inflate_weights(logger)
+            # else:
+                # Directly load 3D model.
+                # load_checkpoint(self, self.pretrained, strict=False, logger=logger)
+        elif self.pretrained is None:
+            self.apply(_init_weights)
+        else:
+            raise TypeError('pretrained must be a str or None')
+
+    def forward(self, x):
+        """Forward function."""
+
+        shortcut = x
+        x = self.patch_embed(x)
+
+        x = self.pos_drop(x)
+        #print('after pos drop',x.shape)
+
+        for layer in self.layers:
+            x = layer(x.contiguous())
+
+
+        x = rearrange(x, 'n c d h w -> n d h w c')
+        #print('after rearrange',x.shape)
+        x = self.norm(x)
+        #print('after norm',x.shape)
+        x = rearrange(x, 'n d h w c -> n c d h w')
+        #print('after rearrange',x.shape)
+
+        x = self.patch_unembed(x)
+
+        x = x + shortcut
+
+        if self.task == 'segment':
+            x = self.segmentation_decode(x)
+
+        else:
+            x = self.upsampler(x)
+
+        return x
+
+
+
+

archs/swinir_rcab.py

@@ -0,0 +1,1296 @@
+# Modified from https://github.com/JingyunLiang/SwinIR
+# SwinIR: Image Restoration Using Swin Transformer, https://arxiv.org/abs/2108.10257
+# Originally Written by Ze Liu, Modified by Jingyun Liang.
+
+import collections.abc
+import math
+import torch
+import torch.nn as nn
+import torch.utils.checkpoint as checkpoint
+from itertools import repeat
+
+# from self_attention_cv import AxialAttentionBlock
+
+from functools import reduce, lru_cache
+from operator import mul
+from einops import rearrange
+import sys
+
+
+def make_layer(basic_block, num_basic_block, **kwarg):
+    """Make layers by stacking the same blocks.
+
+    Args:
+        basic_block (nn.module): nn.module class for basic block.
+        num_basic_block (int): number of blocks.
+
+    Returns:
+        nn.Sequential: Stacked blocks in nn.Sequential.
+    """
+    layers = []
+    for _ in range(num_basic_block):
+        layers.append(basic_block(**kwarg))
+    return nn.Sequential(*layers)
+
+class Upsample(nn.Sequential):
+    """Upsample module.
+
+    Args:
+        scale (int): Scale factor. Supported scales: 2^n and 3.
+        num_feat (int): Channel number of intermediate features.
+    """
+
+    def __init__(self, scale, num_feat):
+        m = []
+        if (scale & (scale - 1)) == 0:  # scale = 2^n
+            for _ in range(int(math.log(scale, 2))):
+                m.append(nn.Conv2d(num_feat, 4 * num_feat, 3, 1, 1))
+                m.append(nn.PixelShuffle(2))
+        elif scale == 3:
+            m.append(nn.Conv2d(num_feat, 9 * num_feat, 3, 1, 1))
+            m.append(nn.PixelShuffle(3))
+        else:
+            raise ValueError(f'scale {scale} is not supported. ' 'Supported scales: 2^n and 3.')
+        super(Upsample, self).__init__(*m)
+
+
+# From PyTorch
+def _ntuple(n):
+
+    def parse(x):
+        if isinstance(x, collections.abc.Iterable):
+            return x
+        return tuple(repeat(x, n))
+
+    return parse
+to_2tuple = _ntuple(2)
+
+def _no_grad_trunc_normal_(tensor, mean, std, a, b):
+    # From: https://github.com/rwightman/pytorch-image-models/blob/master/timm/models/layers/weight_init.py
+    # Cut & paste from PyTorch official master until it's in a few official releases - RW
+    # Method based on https://people.sc.fsu.edu/~jburkardt/presentations/truncated_normal.pdf
+    def norm_cdf(x):
+        # Computes standard normal cumulative distribution function
+        return (1. + math.erf(x / math.sqrt(2.))) / 2.
+
+    if (mean < a - 2 * std) or (mean > b + 2 * std):
+        warnings.warn(
+            'mean is more than 2 std from [a, b] in nn.init.trunc_normal_. '
+            'The distribution of values may be incorrect.',
+            stacklevel=2)
+
+    with torch.no_grad():
+        # Values are generated by using a truncated uniform distribution and
+        # then using the inverse CDF for the normal distribution.
+        # Get upper and lower cdf values
+        low = norm_cdf((a - mean) / std)
+        up = norm_cdf((b - mean) / std)
+
+        # Uniformly fill tensor with values from [low, up], then translate to
+        # [2l-1, 2u-1].
+        tensor.uniform_(2 * low - 1, 2 * up - 1)
+
+        # Use inverse cdf transform for normal distribution to get truncated
+        # standard normal
+        tensor.erfinv_()
+
+        # Transform to proper mean, std
+        tensor.mul_(std * math.sqrt(2.))
+        tensor.add_(mean)
+
+        # Clamp to ensure it's in the proper range
+        tensor.clamp_(min=a, max=b)
+        return tensor
+
+
+def trunc_normal_(tensor, mean=0., std=1., a=-2., b=2.):
+    r"""Fills the input Tensor with values drawn from a truncated
+    normal distribution.
+
+    From: https://github.com/rwightman/pytorch-image-models/blob/master/timm/models/layers/weight_init.py
+
+    The values are effectively drawn from the
+    normal distribution :math:`\mathcal{N}(\text{mean}, \text{std}^2)`
+    with values outside :math:`[a, b]` redrawn until they are within
+    the bounds. The method used for generating the random values works
+    best when :math:`a \leq \text{mean} \leq b`.
+
+    Args:
+        tensor: an n-dimensional `torch.Tensor`
+        mean: the mean of the normal distribution
+        std: the standard deviation of the normal distribution
+        a: the minimum cutoff value
+        b: the maximum cutoff value
+
+    Examples:
+        >>> w = torch.empty(3, 5)
+        >>> nn.init.trunc_normal_(w)
+    """
+    return _no_grad_trunc_normal_(tensor, mean, std, a, b)
+
+class ChannelAttention(nn.Module):
+    """Channel attention used in RCAN.
+
+    Args:
+        num_feat (int): Channel number of intermediate features.
+        squeeze_factor (int): Channel squeeze factor. Default: 16.
+    """
+
+    def __init__(self, num_feat, squeeze_factor=16):
+        super(ChannelAttention, self).__init__()
+        self.attention = nn.Sequential(
+            nn.AdaptiveAvgPool2d(1), nn.Conv2d(num_feat, num_feat // squeeze_factor, 1, padding=0),
+            nn.ReLU(inplace=True), nn.Conv2d(num_feat // squeeze_factor, num_feat, 1, padding=0), nn.Sigmoid())
+
+    def forward(self, x):
+        y = self.attention(x)
+        return x * y
+
+
+class RCAB(nn.Module):
+    """Residual Channel Attention Block (RCAB) used in RCAN.
+
+    Args:
+        num_feat (int): Channel number of intermediate features.
+        squeeze_factor (int): Channel squeeze factor. Default: 16.
+        res_scale (float): Scale the residual. Default: 1.
+    """
+
+    def __init__(self, num_feat, squeeze_factor=16, res_scale=1):
+        super(RCAB, self).__init__()
+        self.res_scale = res_scale
+
+        self.rcab = nn.Sequential(
+            nn.Conv2d(num_feat, num_feat, 3, 1, 1), nn.ReLU(True), nn.Conv2d(num_feat, num_feat, 3, 1, 1),
+            ChannelAttention(num_feat, squeeze_factor))
+
+    def forward(self, x):
+        res = self.rcab(x) * self.res_scale
+        return res + x
+
+
+class ResidualGroup(nn.Module):
+    """Residual Group of RCAB.
+
+    Args:
+        num_feat (int): Channel number of intermediate features.
+        num_block (int): Block number in the body network.
+        squeeze_factor (int): Channel squeeze factor. Default: 16.
+        res_scale (float): Scale the residual. Default: 1.
+    """
+
+    def __init__(self, num_feat, num_block, squeeze_factor=16, res_scale=1):
+        super(ResidualGroup, self).__init__()
+
+        self.residual_group = make_layer(
+            RCAB, num_block, num_feat=num_feat, squeeze_factor=squeeze_factor, res_scale=res_scale)
+        self.conv = nn.Conv2d(num_feat, num_feat, 3, 1, 1)
+
+    def forward(self, x):
+        res = self.conv(self.residual_group(x))
+        return res + x
+
+
+
+
+def drop_path(x, drop_prob: float = 0., training: bool = False):
+    """Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).
+
+    From: https://github.com/rwightman/pytorch-image-models/blob/master/timm/models/layers/drop.py
+    """
+    if drop_prob == 0. or not training:
+        return x
+    keep_prob = 1 - drop_prob
+    shape = (x.shape[0], ) + (1, ) * (x.ndim - 1)  # work with diff dim tensors, not just 2D ConvNets
+    random_tensor = keep_prob + torch.rand(shape, dtype=x.dtype, device=x.device)
+    random_tensor.floor_()  # binarize
+    output = x.div(keep_prob) * random_tensor
+    return output
+
+
+class DropPath(nn.Module):
+    """Drop paths (Stochastic Depth) per sample  (when applied in main path of residual blocks).
+
+    From: https://github.com/rwightman/pytorch-image-models/blob/master/timm/models/layers/drop.py
+    """
+
+    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)
+
+
+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)
+        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):
+    r""" 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):
+        """
+        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
+
+    def extra_repr(self) -> str:
+        return f'dim={self.dim}, window_size={self.window_size}, num_heads={self.num_heads}'
+
+    def flops(self, n):
+        # calculate flops for 1 window with token length of n
+        flops = 0
+        # qkv = self.qkv(x)
+        flops += n * self.dim * 3 * self.dim
+        # attn = (q @ k.transpose(-2, -1))
+        flops += self.num_heads * n * (self.dim // self.num_heads) * n
+        #  x = (attn @ v)
+        flops += self.num_heads * n * n * (self.dim // self.num_heads)
+        # x = self.proj(x)
+        flops += n * self.dim * self.dim
+        return flops
+
+
+class SwinTransformerBlock(nn.Module):
+    r""" Swin Transformer Block.
+
+    Args:
+        dim (int): Number of input channels.
+        input_resolution (tuple[int]): Input resolution.
+        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,
+                 input_resolution,
+                 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.input_resolution = input_resolution
+        self.num_heads = num_heads
+        self.window_size = window_size
+        self.shift_size = shift_size
+        self.mlp_ratio = mlp_ratio
+        if min(self.input_resolution) <= self.window_size:
+            # if window size is larger than input resolution, we don't partition windows
+            self.shift_size = 0
+            self.window_size = min(self.input_resolution)
+        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)
+
+        if self.shift_size > 0:
+            attn_mask = self.calculate_mask(self.input_resolution)
+        else:
+            attn_mask = None
+
+        self.register_buffer('attn_mask', attn_mask)
+
+    def calculate_mask(self, x_size):
+        # calculate attention mask for SW-MSA
+        h, w = x_size
+        img_mask = torch.zeros((1, h, w, 1))  # 1 h w 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))
+
+        return attn_mask
+
+    def forward(self, x, x_size):
+        h, w = x_size
+        b, _, c = x.shape
+        # assert seq_len == h * w, "input feature has wrong size"
+
+        shortcut = x
+        x = self.norm1(x)
+        x = x.view(b, h, w, c)
+
+        # cyclic shift
+        if self.shift_size > 0:
+            shifted_x = torch.roll(x, shifts=(-self.shift_size, -self.shift_size), dims=(1, 2))
+        else:
+            shifted_x = x
+
+        # 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 (to be compatible for testing on images whose shapes are the multiple of window size
+        if self.input_resolution == x_size:
+            attn_windows = self.attn(x_windows, mask=self.attn_mask)  # nw*b, window_size*window_size, c
+        else:
+            attn_windows = self.attn(x_windows, mask=self.calculate_mask(x_size).to(x.device))
+
+        # merge windows
+        attn_windows = attn_windows.view(-1, self.window_size, self.window_size, c)
+        shifted_x = window_reverse(attn_windows, self.window_size, h, w)  # 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
+        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
+
+    def extra_repr(self) -> str:
+        return (f'dim={self.dim}, input_resolution={self.input_resolution}, num_heads={self.num_heads}, '
+                f'window_size={self.window_size}, shift_size={self.shift_size}, mlp_ratio={self.mlp_ratio}')
+
+    def flops(self):
+        flops = 0
+        h, w = self.input_resolution
+        # norm1
+        flops += self.dim * h * w
+        # W-MSA/SW-MSA
+        nw = h * w / self.window_size / self.window_size
+        flops += nw * self.attn.flops(self.window_size * self.window_size)
+        # mlp
+        flops += 2 * h * w * self.dim * self.dim * self.mlp_ratio
+        # norm2
+        flops += self.dim * h * w
+        return flops
+
+
+class PatchMerging(nn.Module):
+    r""" Patch Merging Layer.
+
+    Args:
+        input_resolution (tuple[int]): Resolution of input feature.
+        dim (int): Number of input channels.
+        norm_layer (nn.Module, optional): Normalization layer.  Default: nn.LayerNorm
+    """
+
+    def __init__(self, input_resolution, dim, norm_layer=nn.LayerNorm):
+        super().__init__()
+        self.input_resolution = input_resolution
+        self.dim = dim
+        self.reduction = nn.Linear(4 * dim, 2 * dim, bias=False)
+        self.norm = norm_layer(4 * dim)
+
+    def forward(self, x):
+        """
+        x: b, h*w, c
+        """
+        h, w = self.input_resolution
+        b, seq_len, c = x.shape
+        assert seq_len == h * w, 'input feature has wrong size'
+        assert h % 2 == 0 and w % 2 == 0, f'x size ({h}*{w}) are not even.'
+
+        x = x.view(b, h, w, c)
+
+        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
+
+    def extra_repr(self) -> str:
+        return f'input_resolution={self.input_resolution}, dim={self.dim}'
+
+    def flops(self):
+        h, w = self.input_resolution
+        flops = h * w * self.dim
+        flops += (h // 2) * (w // 2) * 4 * self.dim * 2 * self.dim
+        return flops
+
+
+class BasicLayer(nn.Module):
+    """ A basic Swin Transformer layer for one stage.
+
+    Args:
+        dim (int): Number of input channels.
+        input_resolution (tuple[int]): Input resolution.
+        depth (int): Number of blocks.
+        num_heads (int): Number of attention heads.
+        window_size (int): Local window size.
+        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 | 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,
+                 input_resolution,
+                 depth,
+                 num_heads,
+                 window_size,
+                 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.dim = dim
+        self.input_resolution = input_resolution
+        self.depth = depth
+        self.use_checkpoint = use_checkpoint
+
+        # build blocks
+        self.blocks = nn.ModuleList([
+            SwinTransformerBlock(
+                dim=dim,
+                input_resolution=input_resolution,
+                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(input_resolution, dim=dim, norm_layer=norm_layer)
+        else:
+            self.downsample = None
+
+    def forward(self, x, x_size):
+        for blk in self.blocks:
+            if self.use_checkpoint:
+                x = checkpoint.checkpoint(blk, x)
+            else:
+                x = blk(x, x_size)
+        if self.downsample is not None:
+            x = self.downsample(x)
+        return x
+
+    def extra_repr(self) -> str:
+        return f'dim={self.dim}, input_resolution={self.input_resolution}, depth={self.depth}'
+
+    def flops(self):
+        flops = 0
+        for blk in self.blocks:
+            flops += blk.flops()
+        if self.downsample is not None:
+            flops += self.downsample.flops()
+        return flops
+
+
+class RSTB(nn.Module):
+    """Residual Swin Transformer Block (RSTB).
+
+    Args:
+        dim (int): Number of input channels.
+        input_resolution (tuple[int]): Input resolution.
+        depth (int): Number of blocks.
+        num_heads (int): Number of attention heads.
+        window_size (int): Local window size.
+        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 | 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.
+        img_size: Input image size.
+        patch_size: Patch size.
+        resi_connection: The convolutional block before residual connection.
+    """
+
+    def __init__(self,
+                 dim,
+                 input_resolution,
+                 depth,
+                 num_heads,
+                 window_size,
+                 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,
+                 img_size=224,
+                 patch_size=4,
+                 use_rcab=True,
+                 resi_connection='1conv'):
+        super(RSTB, self).__init__()
+
+        self.dim = dim
+        self.input_resolution = input_resolution
+
+        self.residual_group = BasicLayer(
+            dim=dim,
+            input_resolution=input_resolution,
+            depth=depth,
+            num_heads=num_heads,
+            window_size=window_size,
+            mlp_ratio=mlp_ratio,
+            qkv_bias=qkv_bias,
+            qk_scale=qk_scale,
+            drop=drop,
+            attn_drop=attn_drop,
+            drop_path=drop_path,
+            norm_layer=norm_layer,
+            downsample=downsample,
+            use_checkpoint=use_checkpoint)
+
+        # if resi_connection == '1conv':
+            # # ML-SIM v1 v2 v3 v4 v6 v7 v8
+            # self.conv = nn.Conv2d(dim, dim, 3, 1, 1)
+
+            # # ML-SIM v5
+            # # self.conv = nn.Sequential(
+                # # nn.PixelUnshuffle(2),
+                # # nn.Conv2d(4*dim, 4*dim, 3, 1, 1),
+                # # nn.PixelShuffle(2))
+
+        # elif resi_connection == '3conv':
+            # # to save parameters and memory
+            # self.conv = nn.Sequential(
+                # nn.Conv2d(dim, dim // 4, 3, 1, 1), nn.LeakyReLU(negative_slope=0.2, inplace=True),
+                # nn.Conv2d(dim // 4, dim // 4, 1, 1, 0), nn.LeakyReLU(negative_slope=0.2, inplace=True),
+                # nn.Conv2d(dim // 4, dim, 3, 1, 1))
+
+        self.use_rcab = use_rcab
+
+        self.resi_connection1 = nn.Conv2d(dim, dim, 3, 1, 1)
+        if self.use_rcab:
+            self.resi_connection2 = ResidualGroup(num_feat=dim,squeeze_factor=16,num_block=12)
+            self.resi_connection3 = nn.Conv2d(dim, dim, 3, 1, 1)
+
+        self.patch_embed = PatchEmbed(
+            img_size=img_size, patch_size=patch_size, in_chans=0, embed_dim=dim, norm_layer=None)
+
+        self.patch_unembed = PatchUnEmbed(
+            img_size=img_size, patch_size=patch_size, in_chans=0, embed_dim=dim, norm_layer=None)
+
+    def forward(self, x, x_size):
+        # return self.patch_embed(self.conv(self.patch_unembed(self.residual_group(x, x_size), x_size))) + x
+        shortcut = x
+        x = self.patch_unembed(self.residual_group(x, x_size), x_size)
+        x = self.resi_connection1(x)
+        if self.use_rcab:
+            x = self.resi_connection2(x)
+            x = self.resi_connection3(x)
+        x = self.patch_embed(x) + shortcut
+        return x
+
+    def flops(self):
+        flops = 0
+        flops += self.residual_group.flops()
+        h, w = self.input_resolution
+        flops += h * w * self.dim * self.dim * 9
+        flops += self.patch_embed.flops()
+        flops += self.patch_unembed.flops()
+
+        return flops
+
+
+class PatchEmbed(nn.Module):
+    r""" Image to Patch Embedding
+
+    Args:
+        img_size (int): Image size.  Default: 224.
+        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, img_size=224, patch_size=4, in_chans=3, embed_dim=96, norm_layer=None):
+        super().__init__()
+        img_size = to_2tuple(img_size)
+        patch_size = to_2tuple(patch_size)
+        patches_resolution = [img_size[0] // patch_size[0], img_size[1] // patch_size[1]]
+        self.img_size = img_size
+        self.patch_size = patch_size
+        self.patches_resolution = patches_resolution
+        self.num_patches = patches_resolution[0] * patches_resolution[1]
+
+        self.in_chans = in_chans
+        self.embed_dim = embed_dim
+
+        if norm_layer is not None:
+            self.norm = norm_layer(embed_dim)
+        else:
+            self.norm = None
+
+    def forward(self, x):
+        x = x.flatten(2).transpose(1, 2)  # b Ph*Pw c
+        if self.norm is not None:
+            x = self.norm(x)
+        return x
+
+    def flops(self):
+        flops = 0
+        h, w = self.img_size
+        if self.norm is not None:
+            flops += h * w * self.embed_dim
+        return flops
+
+
+class PatchUnEmbed(nn.Module):
+    r""" Image to Patch Unembedding
+
+    Args:
+        img_size (int): Image size.  Default: 224.
+        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, img_size=224, patch_size=4, in_chans=3, embed_dim=96, norm_layer=None):
+        super().__init__()
+        img_size = to_2tuple(img_size)
+        patch_size = to_2tuple(patch_size)
+        patches_resolution = [img_size[0] // patch_size[0], img_size[1] // patch_size[1]]
+        self.img_size = img_size
+        self.patch_size = patch_size
+        self.patches_resolution = patches_resolution
+        self.num_patches = patches_resolution[0] * patches_resolution[1]
+
+        self.in_chans = in_chans
+        self.embed_dim = embed_dim
+
+    def forward(self, x, x_size):
+        x = x.transpose(1, 2).view(x.shape[0], self.embed_dim, x_size[0], x_size[1])  # b Ph*Pw c
+        return x
+
+    def flops(self):
+        flops = 0
+        return flops
+
+
+class Upsample(nn.Sequential):
+    """Upsample module.
+
+    Args:
+        scale (int): Scale factor. Supported scales: 2^n and 3.
+        num_feat (int): Channel number of intermediate features.
+    """
+
+    def __init__(self, scale, num_feat):
+        m = []
+        if (scale & (scale - 1)) == 0:  # scale = 2^n
+            for _ in range(int(math.log(scale, 2))):
+                m.append(nn.Conv2d(num_feat, 4 * num_feat, 3, 1, 1))
+                m.append(nn.PixelShuffle(2))
+        elif scale == 3:
+            m.append(nn.Conv2d(num_feat, 9 * num_feat, 3, 1, 1))
+            m.append(nn.PixelShuffle(3))
+        else:
+            raise ValueError(f'scale {scale} is not supported. ' 'Supported scales: 2^n and 3.')
+        super(Upsample, self).__init__(*m)
+
+
+class UpsampleOneStep(nn.Sequential):
+    """UpsampleOneStep module (the difference with Upsample is that it always only has 1conv + 1pixelshuffle)
+       Used in lightweight SR to save parameters.
+
+    Args:
+        scale (int): Scale factor. Supported scales: 2^n and 3.
+        num_feat (int): Channel number of intermediate features.
+
+    """
+
+    def __init__(self, scale, num_feat, num_out_ch, input_resolution=None):
+        self.num_feat = num_feat
+        self.input_resolution = input_resolution
+        m = []
+        m.append(nn.Conv2d(num_feat, (scale**2) * num_out_ch, 3, 1, 1))
+        m.append(nn.PixelShuffle(scale))
+        super(UpsampleOneStep, self).__init__(*m)
+
+    def flops(self):
+        h, w = self.input_resolution
+        flops = h * w * self.num_feat * 3 * 9
+        return flops
+
+
+class SwinIR_RCAB(nn.Module):
+    r""" SwinIR
+        A PyTorch impl of : `SwinIR: Image Restoration Using Swin Transformer`, based on Swin Transformer.
+
+    Args:
+        img_size (int | tuple(int)): Input image size. Default 64
+        patch_size (int | tuple(int)): Patch size. Default: 1
+        in_chans (int): Number of input image channels. Default: 3
+        embed_dim (int): Patch embedding dimension. Default: 96
+        depths (tuple(int)): Depth of each Swin Transformer layer.
+        num_heads (tuple(int)): Number of attention heads in different layers.
+        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. Default: None
+        drop_rate (float): Dropout rate. Default: 0
+        attn_drop_rate (float): Attention dropout rate. Default: 0
+        drop_path_rate (float): Stochastic depth rate. Default: 0.1
+        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
+        use_checkpoint (bool): Whether to use checkpointing to save memory. Default: False
+        upscale: Upscale factor. 2/3/4/8 for image SR, 1 for denoising and compress artifact reduction
+        img_range: Image range. 1. or 255.
+        upsampler: The reconstruction reconstruction module. 'pixelshuffle'/'pixelshuffledirect'/'nearest+conv'/None
+        resi_connection: The convolutional block before residual connection. '1conv'/'3conv'
+    """
+
+    def __init__(self,
+                 opt,
+                 img_size=256,
+                 patch_size=1,
+                 in_chans=3,
+                 embed_dim=64,
+                 depths=(6, 6),
+                 num_heads=(8,8),
+                 window_size=4,
+                 mlp_ratio=2.,
+                 qkv_bias=True,
+                 qk_scale=None,
+                 drop_rate=0.,
+                 attn_drop_rate=0.,
+                 drop_path_rate=0.1,
+                 norm_layer=nn.LayerNorm,
+                 ape=False,
+                 patch_norm=True,
+                 use_checkpoint=False,
+                 upscale=2,
+                 img_range=1.,
+                 upsampler='',
+                 resi_connection='1conv',
+                 pixelshuffleFactor=1,
+                 use_rcab=True,
+                 out_chans=1,
+                 vis=False,
+                 **kwargs):
+        super().__init__()
+        num_in_ch = in_chans
+        num_out_ch = out_chans#in_chans
+        num_feat = 64
+        self.img_range = img_range
+        if in_chans == 3:
+            rgb_mean = (0.4488, 0.4371, 0.4040)
+            self.mean = torch.Tensor(rgb_mean).view(1, 3, 1, 1)
+        else:
+            self.mean = torch.zeros(1, 1, 1, 1)
+        self.upscale = upscale
+        self.upsampler = upsampler
+        print('received ',depths,use_rcab)
+
+        # ------------------------- 1, shallow feature extraction ------------------------- #
+        # ML-SIM v1 v2 v3 v6
+        # self.conv_first = nn.Conv2d(num_in_ch, embed_dim, 3, 1, 1)
+
+
+        # ML-SIM v4 v5 v8
+        # print('received pixelshufflefactor',pixelshuffleFactor)
+        self.conv_first = nn.Conv2d(round(pixelshuffleFactor**2*num_in_ch), embed_dim, 3, 1, 1)
+        if pixelshuffleFactor >= 1:
+            self.pixelshuffle_encode = nn.PixelUnshuffle(pixelshuffleFactor)
+            self.pixelshuffle_decode = nn.PixelShuffle(pixelshuffleFactor)
+        else: # e.g. 1/3
+            self.pixelshuffle_encode = nn.PixelShuffle(round(1/pixelshuffleFactor))
+            self.pixelshuffle_decode = nn.PixelUnshuffle(round(1/pixelshuffleFactor))
+
+        # ML-SIM v7
+        # pixelshuffleFactor = kwargs['pixelshuffleFactor']
+        # self.conv_first = nn.Conv2d(round(3*pixelshuffleFactor**2*num_in_ch), embed_dim, 3, 1, 1)
+        # if pixelshuffleFactor > 1:
+            # self.pixelshuffle_encode = nn.PixelUnshuffle(pixelshuffleFactor)
+            # self.pixelshuffle_decode = nn.PixelShuffle(pixelshuffleFactor)
+        # else: # e.g. 1/3
+            # self.pixelshuffle_encode = nn.PixelShuffle(round(1/pixelshuffleFactor))
+            # self.pixelshuffle_decode = nn.PixelUnshuffle(round(1/pixelshuffleFactor))
+
+
+
+
+        # ------------------------- 2, deep feature extraction ------------------------- #
+        self.num_layers = len(depths)
+        self.embed_dim = embed_dim
+        self.ape = ape
+        self.patch_norm = patch_norm
+        self.num_features = embed_dim
+        self.mlp_ratio = mlp_ratio
+
+        # split image into non-overlapping patches
+        self.patch_embed = PatchEmbed(
+            img_size=img_size,
+            patch_size=patch_size,
+            in_chans=embed_dim,
+            embed_dim=embed_dim,
+            norm_layer=norm_layer if self.patch_norm else None)
+        num_patches = self.patch_embed.num_patches
+        patches_resolution = self.patch_embed.patches_resolution
+        self.patches_resolution = patches_resolution
+
+        # merge non-overlapping patches into image
+        self.patch_unembed = PatchUnEmbed(
+            img_size=img_size,
+            patch_size=patch_size,
+            in_chans=embed_dim,
+            embed_dim=embed_dim,
+            norm_layer=norm_layer if self.patch_norm else None)
+
+        # absolute position embedding
+        if self.ape:
+            self.absolute_pos_embed = nn.Parameter(torch.zeros(1, num_patches, embed_dim))
+            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 Residual Swin Transformer blocks (RSTB)
+        self.layers = nn.ModuleList()
+        for i_layer in range(self.num_layers):
+            layer = RSTB(
+                dim=embed_dim,
+                input_resolution=(patches_resolution[0], patches_resolution[1]),
+                depth=depths[i_layer],
+                num_heads=num_heads[i_layer],
+                window_size=window_size,
+                mlp_ratio=self.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])],  # no impact on SR results
+                norm_layer=norm_layer,
+                downsample=None,
+                use_checkpoint=use_checkpoint,
+                img_size=img_size,
+                patch_size=patch_size,
+                use_rcab=use_rcab,
+                resi_connection=resi_connection)
+            self.layers.append(layer)
+        self.norm = norm_layer(self.num_features)
+
+        # build the last conv layer in deep feature extraction
+        if resi_connection == '1conv':
+            self.conv_after_body = nn.Conv2d(embed_dim, embed_dim, 3, 1, 1)
+        elif resi_connection == '3conv':
+            # to save parameters and memory
+            self.conv_after_body = nn.Sequential(
+                nn.Conv2d(embed_dim, embed_dim // 4, 3, 1, 1), nn.LeakyReLU(negative_slope=0.2, inplace=True),
+                nn.Conv2d(embed_dim // 4, embed_dim // 4, 1, 1, 0), nn.LeakyReLU(negative_slope=0.2, inplace=True),
+                nn.Conv2d(embed_dim // 4, embed_dim, 3, 1, 1))
+
+        # ------------------------- 3, high quality image reconstruction ------------------------- #
+        if self.upsampler == 'pixelshuffle':
+            # for classical SR
+            self.conv_before_upsample = nn.Sequential(
+                nn.Conv2d(embed_dim, num_feat, 3, 1, 1), nn.LeakyReLU(inplace=True))
+            self.upsample = Upsample(upscale, num_feat)
+            self.conv_last = nn.Conv2d(num_feat, num_out_ch, 3, 1, 1)
+        elif self.upsampler == 'pixelshuffledirect':
+            # for lightweight SR (to save parameters)
+            self.upsample = UpsampleOneStep(upscale, embed_dim, num_out_ch,
+                                            (patches_resolution[0], patches_resolution[1]))
+        elif self.upsampler == 'nearest+conv':
+            # for real-world SR (less artifacts)
+            assert self.upscale == 4, 'only support x4 now.'
+            self.conv_before_upsample = nn.Sequential(
+                nn.Conv2d(embed_dim, num_feat, 3, 1, 1), nn.LeakyReLU(inplace=True))
+            self.conv_up1 = nn.Conv2d(num_feat, num_feat, 3, 1, 1)
+            self.conv_up2 = nn.Conv2d(num_feat, num_feat, 3, 1, 1)
+            self.conv_hr = nn.Conv2d(num_feat, num_feat, 3, 1, 1)
+            self.conv_last = nn.Conv2d(num_feat, num_out_ch, 3, 1, 1)
+            self.lrelu = nn.LeakyReLU(negative_slope=0.2, inplace=True)
+        else:
+            # for image denoising and JPEG compression artifact reduction
+            # self.conv_last = nn.Conv2d(embed_dim, num_out_ch, 3, 1, 1) # original code
+
+            # ML-SIM v1 v6
+            # self.conv_last = nn.Conv2d(embed_dim, num_in_ch, 3, 1, 1)
+            # self.conv_combine = nn.Conv2d(num_in_ch, num_out_ch, 3, 1, 1)
+
+            # ML-SIM v2,v3
+            # self.conv_last = nn.Conv2d(embed_dim, num_in_ch, 3, 1, 1)
+            # self.conv_combine = nn.Conv2d(num_in_ch, num_out_ch, 3, 1, 1)
+            # self.axial_att_block = AxialAttentionBlock(in_channels=9, dim=256, heads=8)
+
+            # ML-SIM v4 v5
+            # self.conv_last = nn.Conv2d(embed_dim, round(pixelshuffleFactor**2*num_in_ch), 3, 1, 1)
+            # self.conv_combine = nn.Conv2d(num_in_ch, num_out_ch, 3, 1, 1)
+
+            # ML-SIM v7
+            # self.conv_last = nn.Conv2d(embed_dim, round(3*pixelshuffleFactor**2*num_in_ch), 3, 1, 1)
+            # self.conv_combine = nn.Conv2d(3*num_in_ch, num_out_ch, 3, 1, 1)
+
+            # ML-SIM v8
+            self.conv_before_upsample = nn.Sequential(
+                nn.Conv2d(embed_dim, num_feat, 3, 1, 1), nn.LeakyReLU(inplace=True))
+            self.upsample = Upsample(upscale, num_feat)
+            self.conv_last = nn.Conv2d(num_feat, round(pixelshuffleFactor**2*num_in_ch), 3, 1, 1)
+            self.conv_combine = nn.Conv2d(num_in_ch, num_out_ch, 3, 1, 1)
+
+        self.task = opt.task
+        if self.task == 'segment':
+            self.segmentation_decode = nn.Conv2d(num_in_ch, 4, 1)
+        self.vis = vis
+        self.apply(self._init_weights)
+
+    def _init_weights(self, 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)
+
+    @torch.jit.ignore
+    def no_weight_decay(self):
+        return {'absolute_pos_embed'}
+
+    @torch.jit.ignore
+    def no_weight_decay_keywords(self):
+        return {'relative_position_bias_table'}
+
+    def forward_features(self, x):
+        x_size = (x.shape[2], x.shape[3])
+        # print('before patch embed',x.shape)
+        x = self.patch_embed(x)
+        # print('after patch embed',x.shape)
+        if self.ape:
+            x = x + self.absolute_pos_embed
+        x = self.pos_drop(x)
+
+        for idx,layer in enumerate(self.layers):
+            x = layer(x, x_size)
+            if self.vis:
+                x_unembed = self.patch_unembed(x, x_size)
+                torch.save(x_unembed.detach().cpu(),'x_layer_%d.pth' %  idx)
+
+
+        x = self.norm(x)  # b seq_len c
+        # rint('before patch unembed',x.shape)
+        x = self.patch_unembed(x, x_size)
+        # print('before patch unembed',x.shape)
+
+        return x
+
+    def forward(self, x):
+        # print('starting forward',x.shape)
+        self.mean = self.mean.type_as(x)
+        x = (x - self.mean) * self.img_range
+
+        if self.upsampler == 'pixelshuffle':
+            # for classical SR
+            x = self.conv_first(x)
+            x = self.conv_after_body(self.forward_features(x)) + x
+            x = self.conv_before_upsample(x)
+            x = self.conv_last(self.upsample(x))
+        elif self.upsampler == 'pixelshuffledirect':
+            # for lightweight SR
+            x = self.conv_first(x)
+            x = self.conv_after_body(self.forward_features(x)) + x
+            x = self.upsample(x)
+        elif self.upsampler == 'nearest+conv':
+            # for real-world SR
+            x = self.conv_first(x)
+            x = self.conv_after_body(self.forward_features(x)) + x
+            x = self.conv_before_upsample(x)
+            x = self.lrelu(self.conv_up1(torch.nn.functional.interpolate(x, scale_factor=2, mode='nearest')))
+            x = self.lrelu(self.conv_up2(torch.nn.functional.interpolate(x, scale_factor=2, mode='nearest')))
+            x = self.conv_last(self.lrelu(self.conv_hr(x)))
+        else:
+            # for image denoising and JPEG compression artifact reduction
+
+            # ML-SIM v1 v2 v3
+            # x_first = self.conv_first(x)
+            # res = self.conv_after_body(self.forward_features(x_first)) + x_first
+            # res = self.conv_last(res)
+
+            # ML-SIM v1
+            # x = self.conv_combine(x + res)
+
+            # ML-SIM v2
+            # x = self.axial_att_block(x)
+            # x = self.conv_combine(x + res)
+
+            # ML-SIM v3
+            # res = self.axial_att_block(res)
+            # x = self.conv_combine(x + res)
+
+            # ML-SIM v4 v5
+            # x_encoded = self.pixelshuffle_encode(x)
+            # x_first = self.conv_first(x_encoded)
+            # res = self.conv_after_body(self.forward_features(x_first)) + x_first
+            # res = self.conv_last(res)
+            # res_decoded = self.pixelshuffle_decode(res)
+            # x = self.conv_combine(x + res_decoded)
+
+
+            # ML-SIM v6
+            # x_encoded = torch.fft.fft2(x,dim=(-1,-2)).real
+            # x_first = self.conv_first(x_encoded + x)
+            # res = self.conv_after_body(self.forward_features(x_first)) + x_first
+            # res = self.conv_last(res)
+            # x = self.conv_combine(x + res)
+
+            # ML-SIM v7
+            # x_cos = torch.cos(x)
+            # x_sin = torch.sin(x)
+            # x = torch.cat((x,x_cos,x_sin),dim=1)
+            # x_encoded = self.pixelshuffle_encode(x)
+            # x_first = self.conv_first(x_encoded)
+            # res = self.conv_after_body(self.forward_features(x_first)) + x_first
+            # res = self.conv_last(res)
+            # res_decoded = self.pixelshuffle_decode(res)
+            # x = self.conv_combine(x + res_decoded)
+
+            # ML-SIM v8
+            x_encoded = self.pixelshuffle_encode(x)
+
+            # print('after pixelshuffle',x_encoded.shape)
+            x_first = self.conv_first(x_encoded)
+            # print('after conv first',x_first.shape)
+            x_forwardfeat = self.forward_features(x_first)
+            # print('after forward feat',x_forwardfeat.shape)
+            res = self.conv_after_body(x_forwardfeat) + x_first
+            # print('after conv after body',res.shape)
+            x = self.conv_before_upsample(res)
+            # print('after conv before upsample',x.shape)
+            x = self.conv_last(self.upsample(x))
+            # print('after conv last',x.shape)
+
+
+            if self.task == 'segment':
+                x = self.segmentation_decode(x) # assumes pixelshuffle = 1
+            else:
+                res_decoded = self.pixelshuffle_decode(x)
+                # print('after pixel shuffle',res_decoded.shape)
+                x = self.conv_combine(res_decoded)
+                # print('after conv combine',x.shape)
+
+
+        x = x / self.img_range + self.mean
+
+        return x
+
+    def flops(self):
+        flops = 0
+        h, w = self.patches_resolution
+        flops += h * w * 3 * self.embed_dim * 9
+        flops += self.patch_embed.flops()
+        for layer in self.layers:
+            flops += layer.flops()
+        flops += h * w * 3 * self.embed_dim * self.embed_dim
+        flops += self.upsample.flops()
+        return flops
+
+
+if __name__ == '__main__':
+    upscale = 4
+    window_size = 8
+    height = (1024 // upscale // window_size + 1) * window_size
+    width = (720 // upscale // window_size + 1) * window_size
+    model = SwinIR(
+        upscale=2,
+        img_size=(height, width),
+        window_size=window_size,
+        img_range=1.,
+        depths=[6, 6, 6, 6],
+        embed_dim=60,
+        num_heads=[6, 6, 6, 6],
+        mlp_ratio=2,
+        upsampler='pixelshuffledirect')
+    print(model)
+    print(height, width, model.flops() / 1e9)
+
+    x = torch.randn((1, 3, height, width))
+    x = model(x)
+    print(x.shape)
+

requirements.txt

@@ -6,3 +6,5 @@ scikit-image
 opencv-python
 numpy
 matplotlib
+timm
+einops