MetaFormer.py

# (ref) https://github.com/sail-sg/metaformer/tree/main

"""
MetaFormer baselines with CAFormer.
Some implementations are modified from timm (https://github.com/rwightman/pytorch-image-models).
"""
from functools import partial
import torch
import torch.nn as nn
import torch.nn.functional as F

from timm.models.layers import trunc_normal_, DropPath
from timm.models.registry import register_model
from timm.data import IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD
from timm.models.layers.helpers import to_2tuple


def _cfg(url='', **kwargs):
    return {
        'url': url,
        'num_classes': 1000, 'input_size': (3, 224, 224), 'pool_size': None,
        'crop_pct': 1.0, 'interpolation': 'bicubic',
        'mean': IMAGENET_DEFAULT_MEAN, 'std': IMAGENET_DEFAULT_STD, 'classifier': 'head',
        **kwargs
    }


default_cfgs = {
    'caformer_s18': _cfg(url='https://huggingface.co/sail/dl/resolve/main/caformer/caformer_s18.pth'),
    'caformer_s18_384': _cfg(url='https://huggingface.co/sail/dl/resolve/main/caformer/caformer_s18_384.pth',
                            input_size=(3, 384, 384)),
    'caformer_s18_in21ft1k': _cfg(url='https://huggingface.co/sail/dl/resolve/main/caformer/caformer_s18_in21ft1k.pth'),
    'caformer_s18_384_in21ft1k': _cfg(url='https://huggingface.co/sail/dl/resolve/main/caformer/caformer_s18_384_in21ft1k.pth',
                                    input_size=(3, 384, 384)),
    'caformer_s18_in21k': _cfg(url='https://huggingface.co/sail/dl/resolve/main/caformer/caformer_s18_in21k.pth',
                            num_classes=21841),
}


class Downsampling(nn.Module):
    """
    Downsampling (Stem) implemented by a layer of convolution.
    """
    def __init__(self, in_channels, out_channels, 
        kernel_size, stride=1, padding=0, 
        pre_norm=None, post_norm=None, pre_permute=False):
        super().__init__()
        self.pre_norm = pre_norm(in_channels) if pre_norm else nn.Identity()
        self.pre_permute = pre_permute
        self.conv = nn.Conv2d(in_channels, out_channels, kernel_size=kernel_size, 
                            stride=stride, padding=padding)
        self.post_norm = post_norm(out_channels) if post_norm else nn.Identity()

    def forward(self, x):
        x = self.pre_norm(x)
        if self.pre_permute:
            # if take [B, H, W, C] as input, permute it to [B, C, H, W]
            x = x.permute(0, 3, 1, 2)
        x = self.conv(x)
        x = x.permute(0, 2, 3, 1) # [B, C, H, W] -> [B, H, W, C]
        x = self.post_norm(x)
        return x

# ------

class Scale(nn.Module):
    """
    Scale vector by element multiplications.
    """
    def __init__(self, dim, init_value=1.0, trainable=True):
        super().__init__()
        self.scale = nn.Parameter(init_value * torch.ones(dim), requires_grad=trainable)

    def forward(self, x):
        return x * self.scale


class StarReLU(nn.Module):
    """
    StarReLU: s * relu(x) ** 2 + b
    """
    def __init__(self, scale_value=1.0, bias_value=0.0,
        scale_learnable=True, bias_learnable=True, 
        mode=None, inplace=False):
        super().__init__()
        self.inplace = inplace
        self.relu = nn.ReLU(inplace=inplace)
        self.scale = nn.Parameter(scale_value * torch.ones(1),
            requires_grad=scale_learnable)
        self.bias = nn.Parameter(bias_value * torch.ones(1),
            requires_grad=bias_learnable)
    def forward(self, x):
        return self.scale * self.relu(x)**2 + self.bias


class SquaredReLU(nn.Module):
    """
        Squared ReLU: https://arxiv.org/abs/2109.08668
    """
    def __init__(self, inplace=False):
        super().__init__()
        self.relu = nn.ReLU(inplace=inplace)
    def forward(self, x):
        return torch.square(self.relu(x))        


# === Token-Mixer === # 
class Attention(nn.Module):
    """
    Vanilla self-attention from Transformer: https://arxiv.org/abs/1706.03762.
    Modified from timm.
    """
    def __init__(self, dim, head_dim=32, num_heads=None, qkv_bias=False,
        attn_drop=0., proj_drop=0., proj_bias=False, **kwargs):
        super().__init__()

        self.head_dim = head_dim
        self.scale = head_dim ** -0.5

        self.num_heads = num_heads if num_heads else dim // head_dim
        if self.num_heads == 0:
            self.num_heads = 1
        
        self.attention_dim = self.num_heads * self.head_dim

        self.qkv = nn.Linear(dim, self.attention_dim * 3, bias=qkv_bias)
        self.attn_drop = nn.Dropout(attn_drop)
        self.proj = nn.Linear(self.attention_dim, dim, bias=proj_bias)
        self.proj_drop = nn.Dropout(proj_drop)

        
    def forward(self, x):
        B, H, W, C = x.shape
        N = H * W
        qkv = self.qkv(x).reshape(B, N, 3, self.num_heads, self.head_dim).permute(2, 0, 3, 1, 4)
        q, k, v = qkv.unbind(0)   # make torchscript happy (cannot use tensor as tuple)

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

        x = (attn @ v).transpose(1, 2).reshape(B, H, W, self.attention_dim)
        x = self.proj(x)
        x = self.proj_drop(x)
        return x


class SepConv(nn.Module):
    r"""
    Inverted separable convolution from MobileNetV2: https://arxiv.org/abs/1801.04381.
    """
    def __init__(self, dim, expansion_ratio=2,
        act1_layer=StarReLU, act2_layer=nn.Identity, 
        bias=False, kernel_size=7, padding=3,
        **kwargs, ):
        super().__init__()
        med_channels = int(expansion_ratio * dim)
        self.pwconv1 = nn.Linear(dim, med_channels, bias=bias)
        self.act1 = act1_layer()
        self.dwconv = nn.Conv2d(
            med_channels, med_channels, kernel_size=kernel_size,
            padding=padding, groups=med_channels, bias=bias) # depthwise conv
        self.act2 = act2_layer()
        self.pwconv2 = nn.Linear(med_channels, dim, bias=bias)

    def forward(self, x):
        x = self.pwconv1(x)
        x = self.act1(x)
        x = x.permute(0, 3, 1, 2)
        x = self.dwconv(x)
        x = x.permute(0, 2, 3, 1)
        x = self.act2(x)
        x = self.pwconv2(x)
        return x        


# === Normalization === # 

class LayerNormGeneral(nn.Module):
    r""" General LayerNorm for different situations.

    Args:
        affine_shape (int, list or tuple): The shape of affine weight and bias.
            Usually the affine_shape=C, but in some implementation, like torch.nn.LayerNorm,
            the affine_shape is the same as normalized_dim by default. 
            To adapt to different situations, we offer this argument here.
        normalized_dim (tuple or list): Which dims to compute mean and variance. 
        scale (bool): Flag indicates whether to use scale or not.
        bias (bool): Flag indicates whether to use scale or not.

        We give several examples to show how to specify the arguments.

        LayerNorm (https://arxiv.org/abs/1607.06450):
            For input shape of (B, *, C) like (B, N, C) or (B, H, W, C),
                affine_shape=C, normalized_dim=(-1, ), scale=True, bias=True;
            For input shape of (B, C, H, W),
                affine_shape=(C, 1, 1), normalized_dim=(1, ), scale=True, bias=True.

        Modified LayerNorm (https://arxiv.org/abs/2111.11418)
            that is idental to partial(torch.nn.GroupNorm, num_groups=1):
            For input shape of (B, N, C),
                affine_shape=C, normalized_dim=(1, 2), scale=True, bias=True;
            For input shape of (B, H, W, C),
                affine_shape=C, normalized_dim=(1, 2, 3), scale=True, bias=True;
            For input shape of (B, C, H, W),
                affine_shape=(C, 1, 1), normalized_dim=(1, 2, 3), scale=True, bias=True.

        For the several metaformer baslines,
            IdentityFormer, RandFormer and PoolFormerV2 utilize Modified LayerNorm without bias (bias=False);
            ConvFormer and CAFormer utilizes LayerNorm without bias (bias=False).
    """
    def __init__(self, affine_shape=None, normalized_dim=(-1, ), scale=True, 
        bias=True, eps=1e-5):
        super().__init__()
        self.normalized_dim = normalized_dim
        self.use_scale = scale
        self.use_bias = bias
        self.weight = nn.Parameter(torch.ones(affine_shape)) if scale else None
        self.bias = nn.Parameter(torch.zeros(affine_shape)) if bias else None
        self.eps = eps

    def forward(self, x):
        c = x - x.mean(self.normalized_dim, keepdim=True)
        s = c.pow(2).mean(self.normalized_dim, keepdim=True)
        x = c / torch.sqrt(s + self.eps)
        if self.use_scale:
            x = x * self.weight
        if self.use_bias:
            x = x + self.bias
        return x


class LayerNormWithoutBias(nn.Module):
    """
    Equal to partial(LayerNormGeneral, bias=False) but faster, 
    because it directly utilizes otpimized F.layer_norm
    """
    def __init__(self, normalized_shape, eps=1e-5, **kwargs):
        super().__init__()
        self.eps = eps
        self.bias = None
        if isinstance(normalized_shape, int):
            normalized_shape = (normalized_shape,)
        self.weight = nn.Parameter(torch.ones(normalized_shape))
        self.normalized_shape = normalized_shape
    def forward(self, x):
        return F.layer_norm(x, self.normalized_shape, weight=self.weight, bias=self.bias, eps=self.eps)


class Mlp(nn.Module):
    """ MLP as used in MetaFormer models, eg Transformer, MLP-Mixer, PoolFormer, MetaFormer baslines and related networks.
    Mostly copied from timm.
    """
    def __init__(self, dim, mlp_ratio=4, out_features=None, act_layer=StarReLU, drop=0., bias=False, **kwargs):
        super().__init__()
        in_features = dim
        out_features = out_features or in_features
        hidden_features = int(mlp_ratio * in_features)
        drop_probs = to_2tuple(drop)

        self.fc1 = nn.Linear(in_features, hidden_features, bias=bias)
        self.act = act_layer()
        self.drop1 = nn.Dropout(drop_probs[0])
        self.fc2 = nn.Linear(hidden_features, out_features, bias=bias)
        self.drop2 = nn.Dropout(drop_probs[1])

    def forward(self, x):
        x = self.fc1(x)
        x = self.act(x)
        x = self.drop1(x)
        x = self.fc2(x)
        x = self.drop2(x)
        return x


class MlpHead(nn.Module):
    """ MLP classification head
    """
    def __init__(self, dim, num_classes=1000, mlp_ratio=4, act_layer=SquaredReLU,
        norm_layer=nn.LayerNorm, head_dropout=0., bias=True):
        super().__init__()
        hidden_features = int(mlp_ratio * dim)
        self.fc1 = nn.Linear(dim, hidden_features, bias=bias)
        self.act = act_layer()
        self.norm = norm_layer(hidden_features)
        self.fc2 = nn.Linear(hidden_features, num_classes, bias=bias)
        self.head_dropout = nn.Dropout(head_dropout)


    def forward(self, x):
        x = self.fc1(x)
        x = self.act(x)
        x = self.norm(x)
        x = self.head_dropout(x)
        x = self.fc2(x)
        return x

# ---------

class MetaFormerBlock(nn.Module):
    """
    Implementation of one MetaFormer block.
    """
    def __init__(self, dim,
                token_mixer=nn.Identity, mlp=Mlp,
                norm_layer=nn.LayerNorm,
                drop=0., drop_path=0.,
                layer_scale_init_value=None, res_scale_init_value=None
                ):

        super().__init__()

        self.norm1 = norm_layer(dim)
        self.token_mixer = token_mixer(dim=dim, drop=drop)
        self.drop_path1 = DropPath(drop_path) if drop_path > 0. else nn.Identity()
        self.layer_scale1 = Scale(dim=dim, init_value=layer_scale_init_value) \
            if layer_scale_init_value else nn.Identity()
        self.res_scale1 = Scale(dim=dim, init_value=res_scale_init_value) \
            if res_scale_init_value else nn.Identity()

        self.norm2 = norm_layer(dim)
        self.mlp = mlp(dim=dim, drop=drop)
        self.drop_path2 = DropPath(drop_path) if drop_path > 0. else nn.Identity()
        self.layer_scale2 = Scale(dim=dim, init_value=layer_scale_init_value) \
            if layer_scale_init_value else nn.Identity()
        self.res_scale2 = Scale(dim=dim, init_value=res_scale_init_value) \
            if res_scale_init_value else nn.Identity()
        
    def forward(self, x):
        x = self.res_scale1(x) + \
            self.layer_scale1(
                self.drop_path1(
                    self.token_mixer(self.norm1(x))
                )
            )
        x = self.res_scale2(x) + \
            self.layer_scale2(
                self.drop_path2(
                    self.mlp(self.norm2(x))
                )
            )
        return x        


# -------------

r"""
downsampling (stem) for the first stage is a layer of conv with k7, s4 and p2
downsamplings for the last 3 stages is a layer of conv with k3, s2 and p1
DOWNSAMPLE_LAYERS_FOUR_STAGES format: [Downsampling, Downsampling, Downsampling, Downsampling]
use `partial` to specify some arguments
"""
DOWNSAMPLE_LAYERS_FOUR_STAGES = [partial(Downsampling, kernel_size=7, stride=4, padding=2, post_norm=partial(LayerNormGeneral, bias=False, eps=1e-6))] + \
                                [partial(Downsampling, kernel_size=3, stride=2, padding=1, pre_norm=partial(LayerNormGeneral, bias=False, eps=1e-6), pre_permute=True)]*3



class MetaFormer(nn.Module):
    r""" MetaFormer
        A PyTorch impl of : `MetaFormer Baselines for Vision`  -
          https://arxiv.org/abs/2210.13452

    Args:
        in_chans (int): Number of input image channels. Default: 3.
        num_classes (int): Number of classes for classification head. Default: 1000.
        depths (list or tuple): Number of blocks at each stage. Default: [2, 2, 6, 2].
        dims (int): Feature dimension at each stage. Default: [64, 128, 320, 512].
        downsample_layers: (list or tuple): Downsampling layers before each stage.
        token_mixers (list, tuple or token_fcn): Token mixer for each stage. Default: nn.Identity.
        mlps (list, tuple or mlp_fcn): Mlp for each stage. Default: Mlp.
        norm_layers (list, tuple or norm_fcn): Norm layers for each stage. Default: partial(LayerNormGeneral, eps=1e-6, bias=False).
        drop_path_rate (float): Stochastic depth rate. Default: 0.
        head_dropout (float): dropout for MLP classifier. Default: 0.
        layer_scale_init_values (list, tuple, float or None): Init value for Layer Scale. Default: None.
            None means not use the layer scale. Form: https://arxiv.org/abs/2103.17239.
        res_scale_init_values (list, tuple, float or None): Init value for Layer Scale. Default: [None, None, 1.0, 1.0].
            None means not use the layer scale. From: https://arxiv.org/abs/2110.09456.
        output_norm: norm before classifier head. Default: partial(nn.LayerNorm, eps=1e-6).
        head_fn: classification head. Default: nn.Linear.
    """
    def __init__(self, in_chans=3, num_classes=1000, 
                 depths=[2, 2, 6, 2],
                 dims=[64, 128, 320, 512],
                 downsample_layers=DOWNSAMPLE_LAYERS_FOUR_STAGES,
                 token_mixers=nn.Identity,
                 mlps=Mlp,
                 norm_layers=partial(LayerNormWithoutBias, eps=1e-6), # partial(LayerNormGeneral, eps=1e-6, bias=False),
                 drop_path_rate=0.,
                 head_dropout=0.0, 
                 layer_scale_init_values=None,
                 res_scale_init_values=[None, None, 1.0, 1.0],
                 output_norm=partial(nn.LayerNorm, eps=1e-6), 
                 head_fn=nn.Linear,
                 **kwargs,
                 ):
        super().__init__()
        self.num_classes = num_classes

        if not isinstance(depths, (list, tuple)):
            depths = [depths] # it means the model has only one stage
        if not isinstance(dims, (list, tuple)):
            dims = [dims]

        num_stage = len(depths)
        self.num_stage = num_stage

        if not isinstance(downsample_layers, (list, tuple)):
            downsample_layers = [downsample_layers] * num_stage
        down_dims = [in_chans] + dims
        self.downsample_layers = nn.ModuleList(
            [downsample_layers[i](down_dims[i], down_dims[i+1]) for i in range(num_stage)]
        )
        
        if not isinstance(token_mixers, (list, tuple)):
            token_mixers = [token_mixers] * num_stage

        if not isinstance(mlps, (list, tuple)):
            mlps = [mlps] * num_stage

        if not isinstance(norm_layers, (list, tuple)):
            norm_layers = [norm_layers] * num_stage
        
        dp_rates=[x.item() for x in torch.linspace(0, drop_path_rate, sum(depths))]

        if not isinstance(layer_scale_init_values, (list, tuple)):
            layer_scale_init_values = [layer_scale_init_values] * num_stage
        if not isinstance(res_scale_init_values, (list, tuple)):
            res_scale_init_values = [res_scale_init_values] * num_stage

        self.stages = nn.ModuleList() # each stage consists of multiple metaformer blocks
        cur = 0
        for i in range(num_stage):
            stage = nn.Sequential(
                *[MetaFormerBlock(dim=dims[i],
                token_mixer=token_mixers[i],
                mlp=mlps[i],
                norm_layer=norm_layers[i],
                drop_path=dp_rates[cur + j],
                layer_scale_init_value=layer_scale_init_values[i],
                res_scale_init_value=res_scale_init_values[i],
                ) for j in range(depths[i])]
            )
            self.stages.append(stage)
            cur += depths[i]

        self.norm = output_norm(dims[-1])

        if head_dropout > 0.0:
            self.head = head_fn(dims[-1], num_classes, head_dropout=head_dropout)
        else:
            self.head = head_fn(dims[-1], num_classes)

        self.apply(self._init_weights)

    def _init_weights(self, m):
        if isinstance(m, (nn.Conv2d, nn.Linear)):
            trunc_normal_(m.weight, std=.02)
            if m.bias is not None:
                nn.init.constant_(m.bias, 0)

    @torch.jit.ignore
    def no_weight_decay(self):
        return {'norm'}

    def forward_features(self, x):
        for i in range(self.num_stage):
            x = self.downsample_layers[i](x)
            x = self.stages[i](x)
        return self.norm(x.mean([1, 2])) # (B, H, W, C) -> (B, C)

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

# ---------------------

@register_model
def caformer_s18(pretrained=False, **kwargs):
    model = MetaFormer(
        depths=[3, 3, 9, 3],
        dims=[64, 128, 320, 512],
        token_mixers=[SepConv, SepConv, Attention, Attention],
        head_fn=MlpHead,
        **kwargs)
    model.default_cfg = default_cfgs['caformer_s18']
    if pretrained:
        state_dict = torch.hub.load_state_dict_from_url(
            url= model.default_cfg['url'], map_location="cpu", check_hash=True)
        model.load_state_dict(state_dict)
    return model


@register_model
def caformer_s18_384(pretrained=False, **kwargs):
    model = MetaFormer(
        depths=[3, 3, 9, 3],
        dims=[64, 128, 320, 512],
        token_mixers=[SepConv, SepConv, Attention, Attention],
        head_fn=MlpHead,
        **kwargs)
    model.default_cfg = default_cfgs['caformer_s18_384']
    if pretrained:
        state_dict = torch.hub.load_state_dict_from_url(
            url= model.default_cfg['url'], map_location="cpu", check_hash=True)
        model.load_state_dict(state_dict)
    return model    


@register_model
def caformer_s18_in21ft1k(pretrained=False, **kwargs):
    model = MetaFormer(
        depths=[3, 3, 9, 3],
        dims=[64, 128, 320, 512],
        token_mixers=[SepConv, SepConv, Attention, Attention],
        head_fn=MlpHead,
        **kwargs)
    model.default_cfg = default_cfgs['caformer_s18_in21ft1k']
    if pretrained:
        state_dict = torch.hub.load_state_dict_from_url(
            url= model.default_cfg['url'], map_location="cpu", check_hash=True)
        model.load_state_dict(state_dict)
    return model


@register_model
def caformer_s18_384_in21ft1k(pretrained=False, **kwargs):
    model = MetaFormer(
        depths=[3, 3, 9, 3],
        dims=[64, 128, 320, 512],
        token_mixers=[SepConv, SepConv, Attention, Attention],
        head_fn=MlpHead,
        **kwargs)
    model.default_cfg = default_cfgs['caformer_s18_384_in21ft1k']
    if pretrained:
        state_dict = torch.hub.load_state_dict_from_url(
            url= model.default_cfg['url'], map_location="cpu", check_hash=True)
        model.load_state_dict(state_dict)
    return model


@register_model
def caformer_s18_in21k(pretrained=False, **kwargs):
    model = MetaFormer(
        depths=[3, 3, 9, 3],
        dims=[64, 128, 320, 512],
        token_mixers=[SepConv, SepConv, Attention, Attention],
        head_fn=MlpHead,
        **kwargs)
    model.default_cfg = default_cfgs['caformer_s18_in21k']
    if pretrained:
        state_dict = torch.hub.load_state_dict_from_url(
            url= model.default_cfg['url'], map_location="cpu", check_hash=True)
        model.load_state_dict(state_dict)
    return model


if __name__ == "__main__": 
    x = torch.randn(1, 3, 224, 224) # input image 

    model = caformer_s18()
    output = model(x)

    print(output.shape)