KNN_VisionTransformer.py

"""
modified from https://github.com/rwightman/pytorch-image-models/blob/master/timm/models/vision_transformer.py
"""
import math
import logging
from functools import partial
from collections import OrderedDict

import torch
import torch.nn as nn
import torch.nn.functional as F

from timm.data import IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD
from timm.models.helpers import load_pretrained
from timm.models.layers import DropPath, to_2tuple, trunc_normal_
from timm.models.resnet import resnet26d, resnet50d
from timm.models.resnetv2 import ResNetV2, StdConv2dSame
from timm.models.registry import register_model

_logger = logging.getLogger(__name__)

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


class kNNAttention(nn.Module):
    def __init__(self, dim, num_heads=8, qkv_bias=False, qk_scale=None, attn_drop=0., proj_drop=0.,topk=100):
        super().__init__()
        self.num_heads = num_heads
        head_dim = dim // num_heads
        # NOTE scale factor was wrong in my original version, can set manually to be compat with prev weights
        self.scale = qk_scale or head_dim ** -0.5

        self.qkv = nn.Linear(dim, 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.topk = topk

    def forward(self, x):
        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)
        attn = (q @ k.transpose(-2, -1)) * self.scale
        # the core code block
        mask=torch.zeros(B,self.num_heads,N,N,device=x.device,requires_grad=False)
        index=torch.topk(attn,k=self.topk,dim=-1,largest=True)[1]
        mask.scatter_(-1,index,1.)
        attn=torch.where(mask>0,attn,torch.full_like(attn,float('-inf')))
        # end of the core code block

        attn = attn.softmax(dim=-1)
        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 Block(nn.Module):

    def __init__(self, dim, num_heads, mlp_ratio=4., qkv_bias=False, qk_scale=None, drop=0., attn_drop=0.,
                 drop_path=0., act_layer=nn.GELU, norm_layer=nn.LayerNorm):
        super().__init__()
        self.norm1 = norm_layer(dim)
        self.attn = kNNAttention(
            dim, num_heads=num_heads, qkv_bias=qkv_bias, qk_scale=qk_scale, attn_drop=attn_drop, proj_drop=drop)
        # NOTE: drop path for stochastic depth, we shall see if this is better than dropout here
        self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()
        self.norm2 = norm_layer(dim)
        mlp_hidden_dim = int(dim * mlp_ratio)
        self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop)

    def forward(self, x):
        x = x + self.drop_path(self.attn(self.norm1(x)))
        x = x + self.drop_path(self.mlp(self.norm2(x)))
        return x


class PatchEmbed(nn.Module):
    """ Image to Patch Embedding
    """
    def __init__(self, img_size=224, patch_size=16, in_chans=3, embed_dim=768):
        super().__init__()
        img_size = to_2tuple(img_size)
        patch_size = to_2tuple(patch_size)
        num_patches = (img_size[1] // patch_size[1]) * (img_size[0] // patch_size[0])
        self.img_size = img_size
        self.patch_size = patch_size
        self.num_patches = num_patches

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

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


class HybridEmbed(nn.Module):
    """ CNN Feature Map Embedding
    Extract feature map from CNN, flatten, project to embedding dim.
    """
    def __init__(self, backbone, img_size=224, feature_size=None, in_chans=3, embed_dim=768):
        super().__init__()
        assert isinstance(backbone, nn.Module)
        img_size = to_2tuple(img_size)
        self.img_size = img_size
        self.backbone = backbone
        if feature_size is None:
            with torch.no_grad():
                # FIXME this is hacky, but most reliable way of determining the exact dim of the output feature
                # map for all networks, the feature metadata has reliable channel and stride info, but using
                # stride to calc feature dim requires info about padding of each stage that isn't captured.
                training = backbone.training
                if training:
                    backbone.eval()
                o = self.backbone(torch.zeros(1, in_chans, img_size[0], img_size[1]))
                if isinstance(o, (list, tuple)):
                    o = o[-1]  # last feature if backbone outputs list/tuple of features
                feature_size = o.shape[-2:]
                feature_dim = o.shape[1]
                backbone.train(training)
        else:
            feature_size = to_2tuple(feature_size)
            if hasattr(self.backbone, 'feature_info'):
                feature_dim = self.backbone.feature_info.channels()[-1]
            else:
                feature_dim = self.backbone.num_features
        self.num_patches = feature_size[0] * feature_size[1]
        self.proj = nn.Conv2d(feature_dim, embed_dim, 1)

    def forward(self, x):
        x = self.backbone(x)
        if isinstance(x, (list, tuple)):
            x = x[-1]  # last feature if backbone outputs list/tuple of features
        x = self.proj(x).flatten(2).transpose(1, 2)
        return x


class VisionTransformer(nn.Module):
    """ Vision Transformer

    A PyTorch impl of : `An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale`  -
        https://arxiv.org/abs/2010.11929
    """
    def __init__(self, img_size=224, patch_size=16, in_chans=3, num_classes=1000, embed_dim=768, depth=12,
                 num_heads=12, mlp_ratio=4., qkv_bias=True, qk_scale=None, representation_size=None,
                 drop_rate=0., attn_drop_rate=0., drop_path_rate=0., hybrid_backbone=None, norm_layer=None):
        """
        Args:
            img_size (int, tuple): input image size
            patch_size (int, tuple): patch size
            in_chans (int): number of input channels
            num_classes (int): number of classes for classification head
            embed_dim (int): embedding dimension
            depth (int): depth of transformer
            num_heads (int): number of attention heads
            mlp_ratio (int): ratio of mlp hidden dim to embedding dim
            qkv_bias (bool): enable bias for qkv if True
            qk_scale (float): override default qk scale of head_dim ** -0.5 if set
            representation_size (Optional[int]): enable and set representation layer (pre-logits) to this value if set
            drop_rate (float): dropout rate
            attn_drop_rate (float): attention dropout rate
            drop_path_rate (float): stochastic depth rate
            hybrid_backbone (nn.Module): CNN backbone to use in-place of PatchEmbed module
            norm_layer: (nn.Module): normalization layer
        """
        super().__init__()
        self.num_classes = num_classes
        self.num_features = self.embed_dim = embed_dim  # num_features for consistency with other models
        norm_layer = norm_layer or partial(nn.LayerNorm, eps=1e-6)

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

        self.cls_token = nn.Parameter(torch.zeros(1, 1, embed_dim))
        self.pos_embed = nn.Parameter(torch.zeros(1, num_patches + 1, embed_dim))
        self.pos_drop = nn.Dropout(p=drop_rate)

        dpr = [x.item() for x in torch.linspace(0, drop_path_rate, depth)]  # stochastic depth decay rule
        self.blocks = nn.ModuleList([
            Block(
                dim=embed_dim, num_heads=num_heads, mlp_ratio=mlp_ratio, qkv_bias=qkv_bias, qk_scale=qk_scale,
                drop=drop_rate, attn_drop=attn_drop_rate, drop_path=dpr[i], norm_layer=norm_layer)
            for i in range(depth)])
        self.norm = norm_layer(embed_dim)

        # Representation layer
        if representation_size:
            self.num_features = representation_size
            self.pre_logits = nn.Sequential(OrderedDict([
                ('fc', nn.Linear(embed_dim, representation_size)),
                ('act', nn.Tanh())
            ]))
        else:
            self.pre_logits = nn.Identity()

        # Classifier head
        self.head = nn.Linear(self.num_features, num_classes) if num_classes > 0 else nn.Identity()

        trunc_normal_(self.pos_embed, std=.02)
        trunc_normal_(self.cls_token, std=.02)
        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 {'pos_embed', 'cls_token'}

    def get_classifier(self):
        return self.head

    def reset_classifier(self, num_classes, global_pool=''):
        self.num_classes = num_classes
        self.head = nn.Linear(self.embed_dim, num_classes) if num_classes > 0 else nn.Identity()

    def forward_features(self, x):
        B = x.shape[0]
        x = self.patch_embed(x)

        cls_tokens = self.cls_token.expand(B, -1, -1)  # stole cls_tokens impl from Phil Wang, thanks
        x = torch.cat((cls_tokens, x), dim=1)
        x = x + self.pos_embed
        x = self.pos_drop(x)

        for blk in self.blocks:
            x = blk(x)

        x = self.norm(x)[:, 0]
        x = self.pre_logits(x)
        return x

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


class DistilledVisionTransformer(VisionTransformer):
    """ Vision Transformer with distillation token.

    Paper: `Training data-efficient image transformers & distillation through attention` -
        https://arxiv.org/abs/2012.12877

    This impl of distilled ViT is taken from https://github.com/facebookresearch/deit
    """
    def __init__(self, *args, **kwargs):
        super().__init__(*args, **kwargs)
        self.dist_token = nn.Parameter(torch.zeros(1, 1, self.embed_dim))
        num_patches = self.patch_embed.num_patches
        self.pos_embed = nn.Parameter(torch.zeros(1, num_patches + 2, self.embed_dim))
        self.head_dist = nn.Linear(self.embed_dim, self.num_classes) if self.num_classes > 0 else nn.Identity()

        trunc_normal_(self.dist_token, std=.02)
        trunc_normal_(self.pos_embed, std=.02)
        self.head_dist.apply(self._init_weights)

    def forward_features(self, x):
        B = x.shape[0]
        x = self.patch_embed(x)

        cls_tokens = self.cls_token.expand(B, -1, -1)  # stole cls_tokens impl from Phil Wang, thanks
        dist_token = self.dist_token.expand(B, -1, -1)
        x = torch.cat((cls_tokens, dist_token, x), dim=1)

        x = x + self.pos_embed
        x = self.pos_drop(x)

        for blk in self.blocks:
            x = blk(x)

        x = self.norm(x)
        return x[:, 0], x[:, 1]

    def forward(self, x):
        x, x_dist = self.forward_features(x)
        x = self.head(x)
        x_dist = self.head_dist(x_dist)
        if self.training:
            return x, x_dist
        else:
            # during inference, return the average of both classifier predictions
            return (x + x_dist) / 2


def resize_pos_embed(posemb, posemb_new):
    # Rescale the grid of position embeddings when loading from state_dict. Adapted from
    # https://github.com/google-research/vision_transformer/blob/00883dd691c63a6830751563748663526e811cee/vit_jax/checkpoint.py#L224
    _logger.info('Resized position embedding: %s to %s', posemb.shape, posemb_new.shape)
    ntok_new = posemb_new.shape[1]
    if True:
        posemb_tok, posemb_grid = posemb[:, :1], posemb[0, 1:]
        ntok_new -= 1
    else:
        posemb_tok, posemb_grid = posemb[:, :0], posemb[0]
    gs_old = int(math.sqrt(len(posemb_grid)))
    gs_new = int(math.sqrt(ntok_new))
    _logger.info('Position embedding grid-size from %s to %s', gs_old, gs_new)
    posemb_grid = posemb_grid.reshape(1, gs_old, gs_old, -1).permute(0, 3, 1, 2)
    posemb_grid = F.interpolate(posemb_grid, size=(gs_new, gs_new), mode='bilinear')
    posemb_grid = posemb_grid.permute(0, 2, 3, 1).reshape(1, gs_new * gs_new, -1)
    posemb = torch.cat([posemb_tok, posemb_grid], dim=1)
    return posemb


def checkpoint_filter_fn(state_dict, model):
    """ convert patch embedding weight from manual patchify + linear proj to conv"""
    out_dict = {}
    if 'model' in state_dict:
        # For deit models
        state_dict = state_dict['model']
    for k, v in state_dict.items():
        if 'patch_embed.proj.weight' in k and len(v.shape) < 4:
            # For old models that I trained prior to conv based patchification
            O, I, H, W = model.patch_embed.proj.weight.shape
            v = v.reshape(O, -1, H, W)
        elif k == 'pos_embed' and v.shape != model.pos_embed.shape:
            # To resize pos embedding when using model at different size from pretrained weights
            v = resize_pos_embed(v, model.pos_embed)
        out_dict[k] = v
    return out_dict


def _create_vision_transformer(variant, pretrained=False, distilled=False, **kwargs):
    default_cfg = default_cfgs[variant]
    default_num_classes = default_cfg['num_classes']
    default_img_size = default_cfg['input_size'][-1]

    num_classes = kwargs.pop('num_classes', default_num_classes)
    img_size = kwargs.pop('img_size', default_img_size)
    repr_size = kwargs.pop('representation_size', None)
    if repr_size is not None and num_classes != default_num_classes:
        # Remove representation layer if fine-tuning. This may not always be the desired action,
        # but I feel better than doing nothing by default for fine-tuning. Perhaps a better interface?
        _logger.warning("Removing representation layer for fine-tuning.")
        repr_size = None

    model_cls = DistilledVisionTransformer if distilled else VisionTransformer
    model = model_cls(img_size=img_size, num_classes=num_classes, representation_size=repr_size, **kwargs)
    model.default_cfg = default_cfg

    if pretrained:
        load_pretrained(
            model, num_classes=num_classes, in_chans=kwargs.get('in_chans', 3),
            filter_fn=partial(checkpoint_filter_fn, model=model))
    return model


@register_model
def vit_small_patch16_224(pretrained=False, **kwargs):
    """ My custom 'small' ViT model. Depth=8, heads=8= mlp_ratio=3."""
    model_kwargs = dict(
        patch_size=16, embed_dim=768, depth=8, num_heads=8, mlp_ratio=3.,
        qkv_bias=False, norm_layer=nn.LayerNorm, **kwargs)
    if pretrained:
        # NOTE my scale was wrong for original weights, leaving this here until I have better ones for this model
        model_kwargs.setdefault('qk_scale', 768 ** -0.5)
    model = _create_vision_transformer('vit_small_patch16_224', pretrained=pretrained, **model_kwargs)
    return model


@register_model
def vit_base_patch16_224(pretrained=False, **kwargs):
    """ ViT-Base (ViT-B/16) from original paper (https://arxiv.org/abs/2010.11929).
    ImageNet-1k weights fine-tuned from in21k @ 224x224, source https://github.com/google-research/vision_transformer.
    """
    model_kwargs = dict(patch_size=16, embed_dim=768, depth=12, num_heads=12, **kwargs)
    model = _create_vision_transformer('vit_base_patch16_224', pretrained=pretrained, **model_kwargs)
    return model


@register_model
def vit_base_patch32_224(pretrained=False, **kwargs):
    """ ViT-Base (ViT-B/32) from original paper (https://arxiv.org/abs/2010.11929). No pretrained weights.
    """
    model_kwargs = dict(patch_size=32, embed_dim=768, depth=12, num_heads=12, **kwargs)
    model = _create_vision_transformer('vit_base_patch32_224', pretrained=pretrained, **model_kwargs)
    return model


@register_model
def vit_base_patch16_384(pretrained=False, **kwargs):
    """ ViT-Base model (ViT-B/16) from original paper (https://arxiv.org/abs/2010.11929).
    ImageNet-1k weights fine-tuned from in21k @ 384x384, source https://github.com/google-research/vision_transformer.
    """
    model_kwargs = dict(patch_size=16, embed_dim=768, depth=12, num_heads=12, **kwargs)
    model = _create_vision_transformer('vit_base_patch16_384', pretrained=pretrained, **model_kwargs)
    return model


@register_model
def vit_base_patch32_384(pretrained=False, **kwargs):
    """ ViT-Base model (ViT-B/32) from original paper (https://arxiv.org/abs/2010.11929).
    ImageNet-1k weights fine-tuned from in21k @ 384x384, source https://github.com/google-research/vision_transformer.
    """
    model_kwargs = dict(patch_size=32, embed_dim=768, depth=12, num_heads=12, **kwargs)
    model = _create_vision_transformer('vit_base_patch32_384', pretrained=pretrained, **model_kwargs)
    return model


@register_model
def vit_large_patch16_224(pretrained=False, **kwargs):
    """ ViT-Large model (ViT-L/32) from original paper (https://arxiv.org/abs/2010.11929).
    ImageNet-1k weights fine-tuned from in21k @ 224x224, source https://github.com/google-research/vision_transformer.
    """
    model_kwargs = dict(patch_size=16, embed_dim=1024, depth=24, num_heads=16, **kwargs)
    model = _create_vision_transformer('vit_large_patch16_224', pretrained=pretrained, **model_kwargs)
    return model


@register_model
def vit_large_patch32_224(pretrained=False, **kwargs):
    """ ViT-Large model (ViT-L/32) from original paper (https://arxiv.org/abs/2010.11929). No pretrained weights.
    """
    model_kwargs = dict(patch_size=32, embed_dim=1024, depth=24, num_heads=16, **kwargs)
    model = _create_vision_transformer('vit_large_patch32_224', pretrained=pretrained, **model_kwargs)
    return model


@register_model
def vit_large_patch16_384(pretrained=False, **kwargs):
    """ ViT-Large model (ViT-L/16) from original paper (https://arxiv.org/abs/2010.11929).
    ImageNet-1k weights fine-tuned from in21k @ 384x384, source https://github.com/google-research/vision_transformer.
    """
    model_kwargs = dict(patch_size=16, embed_dim=1024, depth=24, num_heads=16, **kwargs)
    model = _create_vision_transformer('vit_large_patch16_384', pretrained=pretrained, **model_kwargs)
    return model


@register_model
def vit_large_patch32_384(pretrained=False, **kwargs):
    """ ViT-Large model (ViT-L/32) from original paper (https://arxiv.org/abs/2010.11929).
    ImageNet-1k weights fine-tuned from in21k @ 384x384, source https://github.com/google-research/vision_transformer.
    """
    model_kwargs = dict(patch_size=32, embed_dim=1024, depth=24, num_heads=16, **kwargs)
    model = _create_vision_transformer('vit_large_patch32_384', pretrained=pretrained, **model_kwargs)
    return model


@register_model
def vit_base_patch16_224_in21k(pretrained=False, **kwargs):
    """ ViT-Base model (ViT-B/16) from original paper (https://arxiv.org/abs/2010.11929).
    ImageNet-21k weights @ 224x224, source https://github.com/google-research/vision_transformer.
    """
    model_kwargs = dict(
        patch_size=16, embed_dim=768, depth=12, num_heads=12, representation_size=768, **kwargs)
    model = _create_vision_transformer('vit_base_patch16_224_in21k', pretrained=pretrained, **model_kwargs)
    return model


@register_model
def vit_base_patch32_224_in21k(pretrained=False, **kwargs):
    """ ViT-Base model (ViT-B/32) from original paper (https://arxiv.org/abs/2010.11929).
    ImageNet-21k weights @ 224x224, source https://github.com/google-research/vision_transformer.
    """
    model_kwargs = dict(
        patch_size=32, embed_dim=768, depth=12, num_heads=12, representation_size=768, **kwargs)
    model = _create_vision_transformer('vit_base_patch32_224_in21k', pretrained=pretrained, **model_kwargs)
    return model


@register_model
def vit_large_patch16_224_in21k(pretrained=False, **kwargs):
    """ ViT-Large model (ViT-L/16) from original paper (https://arxiv.org/abs/2010.11929).
    ImageNet-21k weights @ 224x224, source https://github.com/google-research/vision_transformer.
    """
    model_kwargs = dict(
        patch_size=16, embed_dim=1024, depth=24, num_heads=16, representation_size=1024, **kwargs)
    model = _create_vision_transformer('vit_large_patch16_224_in21k', pretrained=pretrained, **model_kwargs)
    return model


@register_model
def vit_large_patch32_224_in21k(pretrained=False, **kwargs):
    """ ViT-Large model (ViT-L/32) from original paper (https://arxiv.org/abs/2010.11929).
    ImageNet-21k weights @ 224x224, source https://github.com/google-research/vision_transformer.
    """
    model_kwargs = dict(
        patch_size=32, embed_dim=1024, depth=24, num_heads=16, representation_size=1024, **kwargs)
    model = _create_vision_transformer('vit_large_patch32_224_in21k', pretrained=pretrained, **model_kwargs)
    return model


@register_model
def vit_huge_patch14_224_in21k(pretrained=False, **kwargs):
    """ ViT-Huge model (ViT-H/14) from original paper (https://arxiv.org/abs/2010.11929).
    ImageNet-21k weights @ 224x224, source https://github.com/google-research/vision_transformer.
    NOTE: converted weights not currently available, too large for github release hosting.
    """
    model_kwargs = dict(
        patch_size=14, embed_dim=1280, depth=32, num_heads=16, representation_size=1280, **kwargs)
    model = _create_vision_transformer('vit_huge_patch14_224_in21k', pretrained=pretrained, **model_kwargs)
    return model


@register_model
def vit_base_resnet50_224_in21k(pretrained=False, **kwargs):
    """ R50+ViT-B/16 hybrid model from original paper (https://arxiv.org/abs/2010.11929).
    ImageNet-21k weights @ 224x224, source https://github.com/google-research/vision_transformer.
    """
    # create a ResNetV2 w/o pre-activation, that uses StdConv and GroupNorm and has 3 stages, no head
    backbone = ResNetV2(
        layers=(3, 4, 9), num_classes=0, global_pool='', in_chans=kwargs.get('in_chans', 3),
        preact=False, stem_type='same', conv_layer=StdConv2dSame)
    model_kwargs = dict(
        embed_dim=768, depth=12, num_heads=12, hybrid_backbone=backbone,
        representation_size=768, **kwargs)
    model = _create_vision_transformer('vit_base_resnet50_224_in21k', pretrained=pretrained, **model_kwargs)
    return model


@register_model
def vit_base_resnet50_384(pretrained=False, **kwargs):
    """ R50+ViT-B/16 hybrid from original paper (https://arxiv.org/abs/2010.11929).
    ImageNet-1k weights fine-tuned from in21k @ 384x384, source https://github.com/google-research/vision_transformer.
    """
    # create a ResNetV2 w/o pre-activation, that uses StdConv and GroupNorm and has 3 stages, no head
    backbone = ResNetV2(
        layers=(3, 4, 9), num_classes=0, global_pool='', in_chans=kwargs.get('in_chans', 3),
        preact=False, stem_type='same', conv_layer=StdConv2dSame)
    model_kwargs = dict(embed_dim=768, depth=12, num_heads=12, hybrid_backbone=backbone, **kwargs)
    model = _create_vision_transformer('vit_base_resnet50_384', pretrained=pretrained, **model_kwargs)
    return model


@register_model
def vit_small_resnet26d_224(pretrained=False, **kwargs):
    """ Custom ViT small hybrid w/ ResNet26D stride 32. No pretrained weights.
    """
    backbone = resnet26d(pretrained=pretrained, in_chans=kwargs.get('in_chans', 3), features_only=True, out_indices=[4])
    model_kwargs = dict(embed_dim=768, depth=8, num_heads=8, mlp_ratio=3, hybrid_backbone=backbone, **kwargs)
    model = _create_vision_transformer('vit_small_resnet26d_224', pretrained=pretrained, **model_kwargs)
    return model


@register_model
def vit_small_resnet50d_s3_224(pretrained=False, **kwargs):
    """ Custom ViT small hybrid w/ ResNet50D 3-stages, stride 16. No pretrained weights.
    """
    backbone = resnet50d(pretrained=pretrained, in_chans=kwargs.get('in_chans', 3), features_only=True, out_indices=[3])
    model_kwargs = dict(embed_dim=768, depth=8, num_heads=8, mlp_ratio=3, hybrid_backbone=backbone, **kwargs)
    model = _create_vision_transformer('vit_small_resnet50d_s3_224', pretrained=pretrained, **model_kwargs)
    return model


@register_model
def vit_base_resnet26d_224(pretrained=False, **kwargs):
    """ Custom ViT base hybrid w/ ResNet26D stride 32. No pretrained weights.
    """
    backbone = resnet26d(pretrained=pretrained, in_chans=kwargs.get('in_chans', 3), features_only=True, out_indices=[4])
    model_kwargs = dict(embed_dim=768, depth=12, num_heads=12, hybrid_backbone=backbone, **kwargs)
    model = _create_vision_transformer('vit_base_resnet26d_224', pretrained=pretrained, **model_kwargs)
    return model


@register_model
def vit_base_resnet50d_224(pretrained=False, **kwargs):
    """ Custom ViT base hybrid w/ ResNet50D stride 32. No pretrained weights.
    """
    backbone = resnet50d(pretrained=pretrained, in_chans=kwargs.get('in_chans', 3), features_only=True, out_indices=[4])
    model_kwargs = dict(embed_dim=768, depth=12, num_heads=12, hybrid_backbone=backbone, **kwargs)
    model = _create_vision_transformer('vit_base_resnet50d_224', pretrained=pretrained, **model_kwargs)
    return model


@register_model
def vit_deit_tiny_patch16_224(pretrained=False, **kwargs):
    """ DeiT-tiny model @ 224x224 from paper (https://arxiv.org/abs/2012.12877).
    ImageNet-1k weights from https://github.com/facebookresearch/deit.
    """
    model_kwargs = dict(patch_size=16, embed_dim=192, depth=12, num_heads=3, **kwargs)
    model = _create_vision_transformer('vit_deit_tiny_patch16_224', pretrained=pretrained, **model_kwargs)
    return model


@register_model
def vit_deit_small_patch16_224(pretrained=False, **kwargs):
    """ DeiT-small model @ 224x224 from paper (https://arxiv.org/abs/2012.12877).
    ImageNet-1k weights from https://github.com/facebookresearch/deit.
    """
    model_kwargs = dict(patch_size=16, embed_dim=384, depth=12, num_heads=6, **kwargs)
    model = _create_vision_transformer('vit_deit_small_patch16_224', pretrained=pretrained, **model_kwargs)
    return model


@register_model
def vit_deit_base_patch16_224(pretrained=False, **kwargs):
    """ DeiT base model @ 224x224 from paper (https://arxiv.org/abs/2012.12877).
    ImageNet-1k weights from https://github.com/facebookresearch/deit.
    """
    model_kwargs = dict(patch_size=16, embed_dim=768, depth=12, num_heads=12, **kwargs)
    model = _create_vision_transformer('vit_deit_base_patch16_224', pretrained=pretrained, **model_kwargs)
    return model


@register_model
def vit_deit_base_patch16_384(pretrained=False, **kwargs):
    """ DeiT base model @ 384x384 from paper (https://arxiv.org/abs/2012.12877).
    ImageNet-1k weights from https://github.com/facebookresearch/deit.
    """
    model_kwargs = dict(patch_size=16, embed_dim=768, depth=12, num_heads=12, **kwargs)
    model = _create_vision_transformer('vit_deit_base_patch16_384', pretrained=pretrained, **model_kwargs)
    return model


@register_model
def vit_deit_tiny_distilled_patch16_224(pretrained=False, **kwargs):
    """ DeiT-tiny distilled model @ 224x224 from paper (https://arxiv.org/abs/2012.12877).
    ImageNet-1k weights from https://github.com/facebookresearch/deit.
    """
    model_kwargs = dict(patch_size=16, embed_dim=192, depth=12, num_heads=3, **kwargs)
    model = _create_vision_transformer(
        'vit_deit_tiny_distilled_patch16_224', pretrained=pretrained,  distilled=True, **model_kwargs)
    return model


@register_model
def vit_deit_small_distilled_patch16_224(pretrained=False, **kwargs):
    """ DeiT-small distilled model @ 224x224 from paper (https://arxiv.org/abs/2012.12877).
    ImageNet-1k weights from https://github.com/facebookresearch/deit.
    """
    model_kwargs = dict(patch_size=16, embed_dim=384, depth=12, num_heads=6, **kwargs)
    model = _create_vision_transformer(
        'vit_deit_small_distilled_patch16_224', pretrained=pretrained,  distilled=True, **model_kwargs)
    return model


@register_model
def vit_deit_base_distilled_patch16_224(pretrained=False, **kwargs):
    """ DeiT-base distilled model @ 224x224 from paper (https://arxiv.org/abs/2012.12877).
    ImageNet-1k weights from https://github.com/facebookresearch/deit.
    """
    model_kwargs = dict(patch_size=16, embed_dim=768, depth=12, num_heads=12, **kwargs)
    model = _create_vision_transformer(
        'vit_deit_base_distilled_patch16_224', pretrained=pretrained,  distilled=True, **model_kwargs)
    return model


@register_model
def vit_deit_base_distilled_patch16_384(pretrained=False, **kwargs):
    """ DeiT-base distilled model @ 384x384 from paper (https://arxiv.org/abs/2012.12877).
    ImageNet-1k weights from https://github.com/facebookresearch/deit.
    """
    model_kwargs = dict(patch_size=16, embed_dim=768, depth=12, num_heads=12, **kwargs)
    model = _create_vision_transformer(
        'vit_deit_base_distilled_patch16_384', pretrained=pretrained, distilled=True, **model_kwargs)
    return model