crossformer.py

import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.utils.checkpoint as checkpoint
from timm.models.layers import DropPath, to_2tuple, trunc_normal_

NEG_INF = -1000000

class Mlp(nn.Module):
    r"""2-layer MLP"""
    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

class DynamicPosBias(nn.Module):
    r"""DPB module
    
    Use a MLP to predict position bias used in attention.
    """
    def __init__(self, dim, num_heads, residual):
        super().__init__()
        self.residual = residual
        self.num_heads = num_heads
        self.pos_dim = dim // 4
        self.pos_proj = nn.Linear(2, self.pos_dim)
        self.pos1 = nn.Sequential(
            nn.LayerNorm(self.pos_dim),
            nn.ReLU(inplace=True),
            nn.Linear(self.pos_dim, self.pos_dim),
        )
        self.pos2 = nn.Sequential(
            nn.LayerNorm(self.pos_dim),
            nn.ReLU(inplace=True),
            nn.Linear(self.pos_dim, self.pos_dim)
        )
        self.pos3 = nn.Sequential(
            nn.LayerNorm(self.pos_dim),
            nn.ReLU(inplace=True),
            nn.Linear(self.pos_dim, self.num_heads)
        )
    def forward(self, biases):
        if self.residual:
            pos = self.pos_proj(biases) # 2Wh-1 * 2Ww-1, heads
            pos = pos + self.pos1(pos)
            pos = pos + self.pos2(pos)
            pos = self.pos3(pos)
        else:
            pos = self.pos3(self.pos2(self.pos1(self.pos_proj(biases))))
        return pos

    def flops(self, N):
        flops = N * 2 * self.pos_dim
        flops += N * self.pos_dim * self.pos_dim
        flops += N * self.pos_dim * self.pos_dim
        flops += N * self.pos_dim * self.num_heads
        return flops

class Attention(nn.Module):
    r""" Multi-head self attention module with dynamic position bias.

    Args:
        dim (int): Number of input channels.
        group_size (tuple[int]): The height and width of the group.
        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, group_size, num_heads, qkv_bias=True, qk_scale=None, attn_drop=0., proj_drop=0.,
                 position_bias=True):

        super().__init__()
        self.dim = dim
        self.group_size = group_size  # Wh, Ww
        self.num_heads = num_heads
        head_dim = dim // num_heads
        self.scale = qk_scale or head_dim ** -0.5
        self.position_bias = position_bias

        if position_bias:
            self.pos = DynamicPosBias(self.dim // 4, self.num_heads, residual=False)
            
            # generate mother-set
            position_bias_h = torch.arange(1 - self.group_size[0], self.group_size[0])
            position_bias_w = torch.arange(1 - self.group_size[1], self.group_size[1])
            biases = torch.stack(torch.meshgrid([position_bias_h, position_bias_w]))  # 2, 2Wh-1, 2Ww-1
            biases = biases.flatten(1).transpose(0, 1).float()
            self.register_buffer("biases", biases, persistent=False)

            # get pair-wise relative position index for each token inside the group
            coords_h = torch.arange(self.group_size[0])
            coords_w = torch.arange(self.group_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.group_size[0] - 1  # shift to start from 0
            relative_coords[:, :, 1] += self.group_size[1] - 1
            relative_coords[:, :, 0] *= 2 * self.group_size[1] - 1
            relative_position_index = relative_coords.sum(-1)  # Wh*Ww, Wh*Ww
            self.register_buffer("relative_position_index", relative_position_index, persistent=False)

        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)

        self.softmax = nn.Softmax(dim=-1)

    def forward(self, x, mask=None):
        """
        Args:
            x: input features with shape of (num_groups*B, N, C)
            mask: (0/-inf) mask with shape of (num_groups, 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
        # @ stands for matrix multiplication
        attn = (q @ k.transpose(-2, -1))

        if self.position_bias:
            pos = self.pos(self.biases) # 2Wh-1 * 2Ww-1, heads
            # select position bias
            relative_position_bias = pos[self.relative_position_index.view(-1)].view(
                self.group_size[0] * self.group_size[1], self.group_size[0] * self.group_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}, group_size={self.group_size}, num_heads={self.num_heads}'

    def flops(self, N):
        # calculate flops for 1 group 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
        if self.position_bias:
            flops += self.pos.flops(N)
        return flops


class CrossFormerBlock(nn.Module):
    r""" CrossFormer Block.

    Args:
        dim (int): Number of input channels.
        input_resolution (tuple[int]): Input resulotion.
        num_heads (int): Number of attention heads.
        group_size (int): Group size.
        interval (int): Interval for LDA.
        lsda_flag (int): use SDA or LDA, 0 for SDA and 1 for LDA.
        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
        num_patch_size
        impl_type (str): 
        use_extra_conv (bool): Extra convolution layer. Default: True
    """

    def __init__(self, dim, input_resolution, num_heads, group_size=7, interval=8, lsda_flag=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, num_patch_size=1, 
                 pad_type=0, use_extra_conv=True, use_cpe=False, no_mask=False, adaptive_interval=False):
        super().__init__()
        self.dim = dim
        self.input_resolution = input_resolution
        self.num_heads = num_heads
        self.group_size = group_size
        self.interval = interval
        self.lsda_flag = lsda_flag
        self.mlp_ratio = mlp_ratio
        self.num_patch_size = num_patch_size
        self.pad_type = pad_type
        self.use_extra_conv = use_extra_conv
        self.use_cpe = use_cpe
        if min(self.input_resolution) <= self.group_size:
            # if group size is larger than input resolution, we don't partition groups
            self.lsda_flag = 0
            self.group_size = min(self.input_resolution)

        self.norm1 = norm_layer(dim)

        self.attn = Attention(
            dim, group_size=to_2tuple(self.group_size), num_heads=num_heads,
            qkv_bias=qkv_bias, qk_scale=qk_scale, attn_drop=attn_drop, proj_drop=drop,
            position_bias=(not use_cpe))

        if self.use_cpe:
            self.cpe = nn.Conv2d(in_channels=input_resolution[0], out_channels=input_resolution[0], kernel_size=3, padding=1, groups=input_resolution[0])
            self.norm_cpe = norm_layer(dim)

        if adaptive_interval:
            self.interval = int(np.ceil(self.input_resolution[0] / self.group_size))

        if self.use_extra_conv:
            self.ex_kernel = [3, 3]
            padding = (self.ex_kernel[0] - 1) // 2
            self.ex_conv = nn.Conv2d(dim, dim, self.ex_kernel, padding=padding, groups=dim)
            self.ex_ln = norm_layer(dim)

        self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()
        self.norm2 = norm_layer(dim, elementwise_affine=True)
        mlp_hidden_dim = int(dim * mlp_ratio)
        self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop)

        # compute attention mask
        attn_mask = None

        if not no_mask:
            H, W = self.input_resolution

            size_div = self.interval * self.group_size if self.lsda_flag == 1 else self.group_size

            pad_w = (size_div - W % size_div) % size_div
            pad_h = (size_div - H % size_div) % size_div

            if self.pad_type == 0:
                pad_l = pad_t = 0
            else:
                pad_l = pad_w // 2
                pad_t = pad_h // 2
            
            pad_r = pad_w - pad_l
            pad_b = pad_h - pad_t

            Hp = H + pad_h
            Wp = W + pad_w

            mask = torch.zeros((1, Hp, Wp, 1))
            if pad_h > 0:
                mask[:, -pad_b:, :, :] = -1
                mask[:, : pad_t, :, :] = -1
            if pad_w > 0:
                mask[:, :, -pad_r:, :] = -1
                mask[:, :, : pad_l, :] = -1

            if self.lsda_flag == 0: # 0 for SDA
                G = Gh = Gw = self.group_size
                nG = Hp * Wp // G**2
                # attn_mask
                if pad_w > 0 or pad_h > 0:
                    mask = mask.reshape(1, Hp // G, G, Wp // G, G, 1).permute(0, 1, 3, 2, 4, 5).contiguous()
                    mask = mask.reshape(nG, 1, G * G)
                    attn_mask = torch.zeros((nG, G * G, G * G))
                    attn_mask = attn_mask.masked_fill(mask < 0, NEG_INF)
                else:
                    attn_mask = None
            else: # 1 for LDA
                I = self.interval
                G = Gh = Gw = self.group_size
                Rh, Rw = Hp // (Gh * I), Wp // (Gw * I)
                nG = I ** 2 * Rh * Rw
                # attn_mask
                if pad_w > 0 or pad_h > 0:
                    mask = mask.reshape(1, Rh, Gh, I, Rw, Gw, I, 1).permute(0, 1, 4, 3, 6, 2, 5, 7).contiguous()
                    mask = mask.reshape(nG, 1, Gh * Gw)
                    attn_mask = torch.zeros((nG, Gh * Gw, Gh * Gw))
                    attn_mask = attn_mask.masked_fill(mask < 0, NEG_INF)
                else:
                    attn_mask = None

        self.register_buffer("attn_mask", attn_mask, persistent=False)

    def forward(self, x):
        H, W = self.input_resolution
        B, L, C = x.shape
        assert L == H * W, "input feature has wrong size %d, %d, %d" % (L, H, W)

        shortcut = x
        x = self.norm1(x)
        x = x.view(B, H, W, C)

        if self.use_cpe:
            x = x + self.norm_cpe(self.cpe(x))

        # padding
        size_div = self.interval * self.group_size if self.lsda_flag == 1 else self.group_size

        pad_w = (size_div - W % size_div) % size_div
        pad_h = (size_div - H % size_div) % size_div

        if self.pad_type == 0:
            pad_l = pad_t = 0
        else:
            pad_l = pad_w // 2
            pad_t = pad_h // 2
        
        pad_r = pad_w - pad_l
        pad_b = pad_h - pad_t

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

        # group embeddings
        if self.lsda_flag == 0: # 0 for SDA
            G = Gh = Gw = self.group_size
            x = x.reshape(B, Hp // G, G, Wp // G, G, C).permute(0, 1, 3, 2, 4, 5)
            x = x.reshape(B * Hp * Wp // G**2, G**2, C)
        else: # 1 for LDA
            I = self.interval
            G = Gh = Gw = self.group_size
            Rh, Rw = Hp // (Gh * I), Wp // (Gw * I)
            x = x.reshape(B, Rh, Gh, I, Rw, Gw, I, C).permute(0, 1, 4, 3, 6, 2, 5, 7).contiguous()
            x = x.reshape(B * Rh * Rw * I * I, Gh * Gw, C)

        # multi-head self-attention
        x = self.attn(x, mask=self.attn_mask)  # nW*B, G*G, C

        # ungroup embeddings
        if self.lsda_flag == 0:
            x = x.reshape(B, Hp // G, Wp // G, G, G, C).permute(0, 1, 3, 2, 4, 5).contiguous() # B, Hp//G, G, Wp//G, G, C
        else:
            x = x.reshape(B, Rh, Rw, I, I, Gh, Gw, C).permute(0, 1, 5, 3, 2, 6, 4, 7).contiguous() # B, Rh, Gh, I, Rw, Gw, I, C
        x = x.view(B, Hp, Wp, C)

        # remove padding
        if pad_w > 0 or pad_h > 0:
            x = x[:, pad_t:H+pad_t, pad_l:W+pad_l, :].contiguous()
        x = x.view(B, H * W, C)

        # FFN
        x = shortcut + self.drop_path(x)
        x = x + self.drop_path(self.mlp(self.norm2(x)))

        if self.use_extra_conv:
            x = x.view(B, H, W, C).permute(0, 3, 1, 2).contiguous()
            x = self.ex_conv(x)
            x = x.permute(0, 2, 3, 1).view(B, H * W, C).contiguous()
            x = self.ex_ln(x)

        return x

    def extra_repr(self) -> str:
        return f"dim={self.dim}, input_resolution={self.input_resolution}, num_heads={self.num_heads}, " \
               f"group_size={self.group_size}, lsda_flag={self.lsda_flag}, mlp_ratio={self.mlp_ratio}, " \
               f"interval={self.interval}"

    def flops(self):
        flops = 0
        H, W = self.input_resolution
        # norm1
        flops += self.dim * H * W
        # LSDA
        nW = H * W / self.group_size / self.group_size
        flops += nW * self.attn.flops(self.group_size * self.group_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, patch_size=[2], num_input_patch_size=1):
        super().__init__()
        self.input_resolution = input_resolution
        self.dim = dim
        self.reductions = nn.ModuleList()
        self.patch_size = patch_size
        self.norm = norm_layer(dim)

        for i, ps in enumerate(patch_size):
            if i == len(patch_size) - 1:
                out_dim = 2 * dim // 2 ** i
            else:
                out_dim = 2 * dim // 2 ** (i + 1)
            stride = 2
            padding = (ps - stride) // 2
            self.reductions.append(nn.Conv2d(dim, out_dim, kernel_size=ps, 
                                                stride=stride, padding=padding))

    def forward(self, x):
        """
        x: B, H*W, C
        """
        H, W = self.input_resolution
        B, L, C = x.shape
        assert L == H * W, "input feature has wrong size"
        assert H % 2 == 0 and W % 2 == 0, f"x size ({H}*{W}) are not even."

        x = self.norm(x)
        x = x.view(B, H, W, C).permute(0, 3, 1, 2)

        xs = []
        for i in range(len(self.reductions)):
            tmp_x = self.reductions[i](x).flatten(2).transpose(1, 2)
            xs.append(tmp_x)
        x = torch.cat(xs, dim=2)
        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
        for i, ps in enumerate(self.patch_size):
            if i == len(self.patch_size) - 1:
                out_dim = 2 * self.dim // 2 ** i
            else:
                out_dim = 2 * self.dim // 2 ** (i + 1)
            flops += (H // 2) * (W // 2) * ps * ps * out_dim * self.dim
        return flops


class Stage(nn.Module):
    """ CrossFormer blocks 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.
        group_size (int): variable G in the paper, one group has GxG embeddings
        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, group_size, interval,
                 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,
                 patch_size_end=[4], num_patch_size=None, use_cpe=False, pad_type=0, 
                 no_mask=False, adaptive_interval=False, use_acl=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()
        for i in range(depth):
            lsda_flag = 0 if (i % 2 == 0) else 1

            # use extra convolution block every 3 blocks
            use_extra_conv = ((i + 1) % 3 == 0) and (i < depth - 1) and use_acl

            self.blocks.append(CrossFormerBlock(dim=dim, input_resolution=input_resolution,
                                 num_heads=num_heads, group_size=group_size[i], interval=interval,
                                 lsda_flag=lsda_flag,
                                 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,
                                 num_patch_size=num_patch_size,
                                 use_extra_conv=use_extra_conv,
                                 use_cpe=use_cpe,
                                 pad_type=pad_type,
                                 no_mask=no_mask,
                                 adaptive_interval=adaptive_interval
                                 ))

        # patch merging layer
        if downsample is not None:
            self.downsample = downsample(input_resolution, dim=dim, norm_layer=norm_layer, 
                                         patch_size=patch_size_end, num_input_patch_size=num_patch_size)
        else:
            self.downsample = None

    def forward(self, x):
        for blk in self.blocks:
            if self.use_checkpoint:
                x = checkpoint.checkpoint(blk, x)
            else:
                x = blk(x)
        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 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[0] // patch_size[0]]
        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

        self.projs = nn.ModuleList()
        for i, ps in enumerate(patch_size):
            if i == len(patch_size) - 1:
                dim = embed_dim // 2 ** i
            else:
                dim = embed_dim // 2 ** (i + 1)
            stride = patch_size[0]
            padding = (ps - patch_size[0]) // 2
            self.projs.append(nn.Conv2d(in_chans, dim, kernel_size=ps, stride=stride, padding=padding))
        if norm_layer is not None:
            self.norm = norm_layer(embed_dim)
        else:
            self.norm = None

    def forward(self, x):
        B, C, H, W = x.shape
        # FIXME look at relaxing size constraints
        assert H == self.img_size[0] and W == self.img_size[1], \
            f"Input image size ({H}*{W}) doesn't match model ({self.img_size[0]}*{self.img_size[1]})."
        xs = []
        for i in range(len(self.projs)):
            tx = self.projs[i](x).flatten(2).transpose(1, 2)
            xs.append(tx)  # B Ph*Pw C
        x = torch.cat(xs, dim=2)
        if self.norm is not None:
            x = self.norm(x)
        return x

    def flops(self):
        Ho, Wo = self.patches_resolution
        flops = 0
        for i, ps in enumerate(self.patch_size):
            if i == len(self.patch_size) - 1:
                dim = self.embed_dim // 2 ** i
            else:
                dim = self.embed_dim // 2 ** (i + 1)
            flops += Ho * Wo * dim * self.in_chans * (self.patch_size[i] * self.patch_size[i])
        if self.norm is not None:
            flops += Ho * Wo * self.embed_dim
        return flops


class CrossFormer(nn.Module):
    r""" CrossFormer
        A PyTorch impl of : `CrossFormer: A Versatile Vision Transformer Based on Cross-scale Attention`  -

    Args:
        img_size (int | tuple(int)): Input image size. Default 224
        patch_size (int | tuple(int)): Patch size. Default: 4
        in_chans (int): Number of input image channels. Default: 3
        num_classes (int): Number of classes for classification head. Default: 1000
        embed_dim (int): Patch embedding dimension. Default: 96
        depths (tuple(int)): Depth of each stage.
        num_heads (tuple(int)): Number of attention heads in different layers.
        group_size (int): Group 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
        use_cpe (bool): Whether to use conditional positional encoding. Default: False
        group_type (str): Strategy to change the group size in different stages. Default: constant
        pad_type (bool): 0 to pad in one direction, otherwise 1. Default: 0
    """

    def __init__(self, img_size=224, patch_size=[4], in_chans=3, num_classes=1000,
                 embed_dim=96, depths=[2, 2, 6, 2], num_heads=[3, 6, 12, 24],
                 group_size=[7, 7, 7, 7], crs_interval=[8, 4, 2, 1], mlp_ratio=4., 
                 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, merge_size=[[2], [2], [2]], use_cpe=False,
                 group_type='constant', pad_type=0, no_mask=False,
                 adaptive_interval=False, use_acl=False,  **kwargs):
        super().__init__()

        self.num_classes = num_classes
        self.num_layers = len(depths)
        self.embed_dim = embed_dim
        self.ape = ape
        self.patch_norm = patch_norm
        self.num_features = int(embed_dim * 2 ** (self.num_layers - 1))
        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=in_chans, 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

        # 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

        # compute group size for each layer
        group_size = self.compute_group_size(group_size, depths, patches_resolution, group_type)

        # build layers
        self.layers = nn.ModuleList()

        num_patch_sizes = [len(patch_size)] + [len(m) for m in merge_size]
        for i_layer in range(self.num_layers):
            patch_size_end = merge_size[i_layer] if i_layer < self.num_layers - 1 else None
            num_patch_size = num_patch_sizes[i_layer]
            layer = Stage(dim=int(embed_dim * 2 ** i_layer),
                               input_resolution=(patches_resolution[0] // (2 ** i_layer),
                                                 patches_resolution[1] // (2 ** i_layer)),
                               depth=depths[i_layer],
                               num_heads=num_heads[i_layer],
                               group_size=group_size[i_layer],
                               interval=crs_interval[i_layer],
                               mlp_ratio=self.mlp_ratio[i_layer],
                               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,
                               patch_size_end=patch_size_end,
                               num_patch_size=num_patch_size,
                               use_cpe=use_cpe,
                               pad_type=pad_type,
                               no_mask=no_mask,
                               adaptive_interval=adaptive_interval,
                               use_acl=use_acl)
            self.layers.append(layer)

        self.norm = norm_layer(self.num_features)
        self.avgpool = nn.AdaptiveAvgPool1d(1)
        self.head = nn.Linear(self.num_features, num_classes) if num_classes > 0 else nn.Identity()

        self.apply(self._init_weights)

    def compute_group_size(self, group_size=[7, 7, 7, 7], depths=[2, 2, 6, 2], resolution=[56, 56], group_type='constant'):
        r"""genenrate group size for crossformer
        
        output:
            - rst_group_size: should be in the shape [[4], [4, 4], [14, 14, 14], [7, 7]] if the depths = [1, 2, 3, 2]
        """
        rst_group_size = []

        # compute linear fraction patch size
        min_size = 4
        total_depth = sum(depths)
        step_size = (1 - min_size / resolution[0]) / total_depth
        group_fraction = np.arange(min_size / resolution[0], 1.0, step_size)
    
        cnt = 0
        for i_stage in range(len(depths)):
            rst_group_size.append([])
            cur_resolution = resolution[0] // 2 ** i_stage
            for i_block in range(depths[i_stage]):
                if group_type == 'constant':
                    # constant group size for each stage
                    rst_group_size[i_stage].append(group_size[i_stage])
                elif group_type == 'linear':
                    # the fraction of group size relative to input resolution grow in linear
                    gz = cur_resolution * group_fraction[cnt]
                    rst_group_size[i_stage].append(max(4, int(np.ceil(gz))))
                elif group_type == 'linear_div':
                    # if fraction > 1/2, let fraction = 1/2 if fraction < 3/4 else 1
                    gz = cur_resolution * group_fraction[cnt]
                    if gz > cur_resolution // 2:
                        gz = cur_resolution if gz > cur_resolution * 3 / 4 or i_stage != 2 else cur_resolution // 2
                    rst_group_size[i_stage].append(max(4, int(np.ceil(gz))))
                elif group_type == 'alter':
                    # if fraction > 1/2, let fraction alter between 1/2 and 1
                    gz = cur_resolution * group_fraction[cnt]
                    if gz > cur_resolution // 2:
                        gz = cur_resolution if cnt % 2 != 0 or i_stage != 2 else cur_resolution // 2
                    rst_group_size[i_stage].append(max(4, int(np.ceil(gz))))
                elif group_type == '7_14':
                    rst_group_size[i_stage].append(group_size[i_stage] if i_stage != 2 or i_block >= 4 else group_size[i_stage] // 2) 
                cnt += 1

        print("Group Size:")
        print(rst_group_size)

        return rst_group_size

    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 = self.patch_embed(x)
        if self.ape:
            x = x + self.absolute_pos_embed
        x = self.pos_drop(x)

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

        x = self.norm(x)  # B L C
        x = self.avgpool(x.transpose(1, 2))  # B C 1
        x = torch.flatten(x, 1)
        return x

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

    def flops(self):
        flops = 0
        flops += self.patch_embed.flops()
        for i, layer in enumerate(self.layers):
            flops += layer.flops()
        flops += self.num_features * self.patches_resolution[0] * self.patches_resolution[1] // (2 ** self.num_layers)
        flops += self.num_features * self.num_classes
        return flops