Add levit, levit_c, and visformer model defs. Largely untested and not finished cleanup.

tests/test_models.py

@@ -15,7 +15,7 @@ if hasattr(torch._C, '_jit_set_profiling_executor'):
     torch._C._jit_set_profiling_mode(False)
 
 # transformer models don't support many of the spatial / feature based model functionalities
-NON_STD_FILTERS = ['vit_*', 'tnt_*', 'pit_*', 'swin_*', 'coat_*', 'cait_*', 'mixer_*']
+NON_STD_FILTERS = ['vit_*', 'tnt_*', 'pit_*', 'swin_*', 'coat_*', 'cait_*', 'mixer_*', 'levit*', 'visformer*']
 NUM_NON_STD = len(NON_STD_FILTERS)
 
 # exclude models that cause specific test failures

timm/models/__init__.py

@@ -15,6 +15,8 @@ from .hrnet import *
 from .inception_resnet_v2 import *
 from .inception_v3 import *
 from .inception_v4 import *
+from .levitc import *
+from .levit import *
 from .mlp_mixer import *
 from .mobilenetv3 import *
 from .nasnet import *
@@ -34,6 +36,7 @@ from .swin_transformer import *
 from .tnt import *
 from .tresnet import *
 from .vgg import *
+from .visformer import *
 from .vision_transformer import *
 from .vision_transformer_hybrid import *
 from .vovnet import *

timm/models/layers/patch_embed.py

@@ -15,7 +15,7 @@ from .helpers import to_2tuple
 class PatchEmbed(nn.Module):
     """ 2D Image to Patch Embedding
     """
-    def __init__(self, img_size=224, patch_size=16, in_chans=3, embed_dim=768, norm_layer=None):
+    def __init__(self, img_size=224, patch_size=16, in_chans=3, embed_dim=768, norm_layer=None, flatten=True):
         super().__init__()
         img_size = to_2tuple(img_size)
         patch_size = to_2tuple(patch_size)
@@ -23,6 +23,7 @@ class PatchEmbed(nn.Module):
         self.patch_size = patch_size
         self.grid_size = (img_size[0] // patch_size[0], img_size[1] // patch_size[1])
         self.num_patches = self.grid_size[0] * self.grid_size[1]
+        self.flatten = flatten
 
         self.proj = nn.Conv2d(in_chans, embed_dim, kernel_size=patch_size, stride=patch_size)
         self.norm = norm_layer(embed_dim) if norm_layer else nn.Identity()
@@ -31,6 +32,8 @@ class PatchEmbed(nn.Module):
         B, C, H, W = x.shape
         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)
+        x = self.proj(x)
+        if self.flatten:
+            x = x.flatten(2).transpose(1, 2)  # BCHW -> BNC
         x = self.norm(x)
         return x

timm/models/levit.py

@@ -0,0 +1,440 @@
+# Copyright (c) 2015-present, Facebook, Inc.
+# All rights reserved.
+
+# Modified from
+# https://github.com/rwightman/pytorch-image-models/blob/master/timm/models/vision_transformer.py
+# Copyright 2020 Ross Wightman, Apache-2.0 License
+import itertools
+
+import torch
+
+from timm.data import IMAGENET_DEFAULT_STD, IMAGENET_DEFAULT_MEAN
+from .vision_transformer import trunc_normal_
+from .registry import register_model
+
+
+def _cfg(url='', **kwargs):
+    return {
+        'url': url,
+        'num_classes': 1000, 'input_size': (3, 224, 224), 'pool_size': None,
+        'crop_pct': .9, 'interpolation': 'bicubic', 'fixed_input_size': True,
+        'mean': IMAGENET_DEFAULT_MEAN, 'std': IMAGENET_DEFAULT_STD,
+        'first_conv': 'patch_embed.proj', 'classifier': 'head',
+        **kwargs
+    }
+
+
+specification = {
+    'levit_128s': {
+        'C': '128_256_384', 'D': 16, 'N': '4_6_8', 'X': '2_3_4', 'drop_path': 0,
+        'weights': 'https://dl.fbaipublicfiles.com/LeViT/LeViT-128S-96703c44.pth'},
+    'levit_128': {
+        'C': '128_256_384', 'D': 16, 'N': '4_8_12', 'X': '4_4_4', 'drop_path': 0,
+        'weights': 'https://dl.fbaipublicfiles.com/LeViT/LeViT-128-b88c2750.pth'},
+    'levit_192': {
+        'C': '192_288_384', 'D': 32, 'N': '3_5_6', 'X': '4_4_4', 'drop_path': 0,
+        'weights': 'https://dl.fbaipublicfiles.com/LeViT/LeViT-192-92712e41.pth'},
+    'levit_256': {
+        'C': '256_384_512', 'D': 32, 'N': '4_6_8', 'X': '4_4_4', 'drop_path': 0,
+        'weights': 'https://dl.fbaipublicfiles.com/LeViT/LeViT-256-13b5763e.pth'},
+    'levit_384': {
+        'C': '384_512_768', 'D': 32, 'N': '6_9_12', 'X': '4_4_4', 'drop_path': 0.1,
+        'weights': 'https://dl.fbaipublicfiles.com/LeViT/LeViT-384-9bdaf2e2.pth'},
+}
+
+__all__ = ['Levit']
+
+
+@register_model
+def levit_128s(num_classes=1000, distillation=True, pretrained=False, fuse=False, **kwargs):
+    return model_factory(**specification['levit_128s'], num_classes=num_classes,
+                         distillation=distillation, pretrained=pretrained, fuse=fuse)
+
+
+@register_model
+def levit_128(num_classes=1000, distillation=True, pretrained=False, fuse=False, **kwargs):
+    return model_factory(**specification['levit_128'], num_classes=num_classes,
+                         distillation=distillation, pretrained=pretrained, fuse=fuse)
+
+
+@register_model
+def levit_192(num_classes=1000, distillation=True, pretrained=False, fuse=False, **kwargs):
+    return model_factory(**specification['levit_192'], num_classes=num_classes,
+                         distillation=distillation, pretrained=pretrained, fuse=fuse)
+
+
+@register_model
+def levit_256(num_classes=1000, distillation=True, pretrained=False, fuse=False, **kwargs):
+    return model_factory(**specification['levit_256'], num_classes=num_classes,
+                         distillation=distillation, pretrained=pretrained, fuse=fuse)
+
+
+@register_model
+def levit_384(num_classes=1000, distillation=True, pretrained=False, fuse=False, **kwargs):
+    return model_factory(**specification['levit_384'], num_classes=num_classes,
+                         distillation=distillation, pretrained=pretrained, fuse=fuse)
+
+
+class ConvNorm(torch.nn.Sequential):
+    def __init__(
+            self, a, b, ks=1, stride=1, pad=0, dilation=1, groups=1, bn_weight_init=1, resolution=-10000):
+        super().__init__()
+        self.add_module('c', torch.nn.Conv2d(a, b, ks, stride, pad, dilation, groups, bias=False))
+        bn = torch.nn.BatchNorm2d(b)
+        torch.nn.init.constant_(bn.weight, bn_weight_init)
+        torch.nn.init.constant_(bn.bias, 0)
+        self.add_module('bn', bn)
+
+    @torch.no_grad()
+    def fuse(self):
+        c, bn = self._modules.values()
+        w = bn.weight / (bn.running_var + bn.eps) ** 0.5
+        w = c.weight * w[:, None, None, None]
+        b = bn.bias - bn.running_mean * bn.weight / (bn.running_var + bn.eps) ** 0.5
+        m = torch.nn.Conv2d(
+            w.size(1), w.size(0), w.shape[2:], stride=self.c.stride,
+            padding=self.c.padding, dilation=self.c.dilation, groups=self.c.groups)
+        m.weight.data.copy_(w)
+        m.bias.data.copy_(b)
+        return m
+
+
+class LinearNorm(torch.nn.Sequential):
+    def __init__(self, a, b, bn_weight_init=1, resolution=-100000):
+        super().__init__()
+        self.add_module('c', torch.nn.Linear(a, b, bias=False))
+        bn = torch.nn.BatchNorm1d(b)
+        torch.nn.init.constant_(bn.weight, bn_weight_init)
+        torch.nn.init.constant_(bn.bias, 0)
+        self.add_module('bn', bn)
+
+    @torch.no_grad()
+    def fuse(self):
+        l, bn = self._modules.values()
+        w = bn.weight / (bn.running_var + bn.eps) ** 0.5
+        w = l.weight * w[:, None]
+        b = bn.bias - bn.running_mean * bn.weight / (bn.running_var + bn.eps) ** 0.5
+        m = torch.nn.Linear(w.size(1), w.size(0))
+        m.weight.data.copy_(w)
+        m.bias.data.copy_(b)
+        return m
+
+    def forward(self, x):
+        l, bn = self._modules.values()
+        x = l(x)
+        return bn(x.flatten(0, 1)).reshape_as(x)
+
+
+class NormLinear(torch.nn.Sequential):
+    def __init__(self, a, b, bias=True, std=0.02):
+        super().__init__()
+        self.add_module('bn', torch.nn.BatchNorm1d(a))
+        l = torch.nn.Linear(a, b, bias=bias)
+        trunc_normal_(l.weight, std=std)
+        if bias:
+            torch.nn.init.constant_(l.bias, 0)
+        self.add_module('l', l)
+
+    @torch.no_grad()
+    def fuse(self):
+        bn, l = self._modules.values()
+        w = bn.weight / (bn.running_var + bn.eps) ** 0.5
+        b = bn.bias - self.bn.running_mean * self.bn.weight / (bn.running_var + bn.eps) ** 0.5
+        w = l.weight * w[None, :]
+        if l.bias is None:
+            b = b @ self.l.weight.T
+        else:
+            b = (l.weight @ b[:, None]).view(-1) + self.l.bias
+        m = torch.nn.Linear(w.size(1), w.size(0))
+        m.weight.data.copy_(w)
+        m.bias.data.copy_(b)
+        return m
+
+
+def b16(n, activation, resolution=224):
+    return torch.nn.Sequential(
+        ConvNorm(3, n // 8, 3, 2, 1, resolution=resolution),
+        activation(),
+        ConvNorm(n // 8, n // 4, 3, 2, 1, resolution=resolution // 2),
+        activation(),
+        ConvNorm(n // 4, n // 2, 3, 2, 1, resolution=resolution // 4),
+        activation(),
+        ConvNorm(n // 2, n, 3, 2, 1, resolution=resolution // 8))
+
+
+class Residual(torch.nn.Module):
+    def __init__(self, m, drop):
+        super().__init__()
+        self.m = m
+        self.drop = drop
+
+    def forward(self, x):
+        if self.training and self.drop > 0:
+            return x + self.m(x) * torch.rand(
+                x.size(0), 1, 1, device=x.device).ge_(self.drop).div(1 - self.drop).detach()
+        else:
+            return x + self.m(x)
+
+
+class Attention(torch.nn.Module):
+    def __init__(
+            self, dim, key_dim, num_heads=8, attn_ratio=4, act_layer=None, resolution=14):
+        super().__init__()
+        self.num_heads = num_heads
+        self.scale = key_dim ** -0.5
+        self.key_dim = key_dim
+        self.nh_kd = nh_kd = key_dim * num_heads
+        self.d = int(attn_ratio * key_dim)
+        self.dh = int(attn_ratio * key_dim) * num_heads
+        self.attn_ratio = attn_ratio
+        h = self.dh + nh_kd * 2
+        self.qkv = LinearNorm(dim, h, resolution=resolution)
+        self.proj = torch.nn.Sequential(
+            act_layer(),
+            LinearNorm(self.dh, dim, bn_weight_init=0, resolution=resolution))
+
+        points = list(itertools.product(range(resolution), range(resolution)))
+        N = len(points)
+        attention_offsets = {}
+        idxs = []
+        for p1 in points:
+            for p2 in points:
+                offset = (abs(p1[0] - p2[0]), abs(p1[1] - p2[1]))
+                if offset not in attention_offsets:
+                    attention_offsets[offset] = len(attention_offsets)
+                idxs.append(attention_offsets[offset])
+        self.attention_biases = torch.nn.Parameter(torch.zeros(num_heads, len(attention_offsets)))
+        self.register_buffer('attention_bias_idxs', torch.LongTensor(idxs).view(N, N))
+
+    @torch.no_grad()
+    def train(self, mode=True):
+        super().train(mode)
+        if mode and hasattr(self, 'ab'):
+            del self.ab
+        else:
+            self.ab = self.attention_biases[:, self.attention_bias_idxs]
+
+    def forward(self, x):  # x (B,N,C)
+        B, N, C = x.shape
+        qkv = self.qkv(x)
+        q, k, v = qkv.view(B, N, self.num_heads, -1).split([self.key_dim, self.key_dim, self.d], dim=3)
+        q = q.permute(0, 2, 1, 3)
+        k = k.permute(0, 2, 1, 3)
+        v = v.permute(0, 2, 1, 3)
+
+        ab = self.attention_biases[:, self.attention_bias_idxs] if self.training else self.ab
+        attn = q @ k.transpose(-2, -1) * self.scale + ab
+
+        attn = attn.softmax(dim=-1)
+        x = (attn @ v).transpose(1, 2).reshape(B, N, self.dh)
+        x = self.proj(x)
+        return x
+
+
+class Subsample(torch.nn.Module):
+    def __init__(self, stride, resolution):
+        super().__init__()
+        self.stride = stride
+        self.resolution = resolution
+
+    def forward(self, x):
+        B, N, C = x.shape
+        x = x.view(B, self.resolution, self.resolution, C)[:, ::self.stride, ::self.stride]
+        return x.reshape(B, -1, C)
+
+
+class AttentionSubsample(torch.nn.Module):
+    def __init__(self, in_dim, out_dim, key_dim, num_heads=8,
+                 attn_ratio=2, act_layer=None, stride=2, resolution=14, resolution_=7):
+        super().__init__()
+        self.num_heads = num_heads
+        self.scale = key_dim ** -0.5
+        self.key_dim = key_dim
+        self.nh_kd = nh_kd = key_dim * num_heads
+        self.d = int(attn_ratio * key_dim)
+        self.dh = int(attn_ratio * key_dim) * self.num_heads
+        self.attn_ratio = attn_ratio
+        self.resolution_ = resolution_
+        self.resolution_2 = resolution_ ** 2
+        h = self.dh + nh_kd
+        self.kv = LinearNorm(in_dim, h, resolution=resolution)
+
+        self.q = torch.nn.Sequential(
+            Subsample(stride, resolution),
+            LinearNorm(in_dim, nh_kd, resolution=resolution_))
+        self.proj = torch.nn.Sequential(
+            act_layer(),
+            LinearNorm(self.dh, out_dim, resolution=resolution_))
+
+        self.stride = stride
+        self.resolution = resolution
+        points = list(itertools.product(range(resolution), range(resolution)))
+        points_ = list(itertools.product(range(resolution_), range(resolution_)))
+        N = len(points)
+        N_ = len(points_)
+        attention_offsets = {}
+        idxs = []
+        for p1 in points_:
+            for p2 in points:
+                size = 1
+                offset = (
+                    abs(p1[0] * stride - p2[0] + (size - 1) / 2),
+                    abs(p1[1] * stride - p2[1] + (size - 1) / 2))
+                if offset not in attention_offsets:
+                    attention_offsets[offset] = len(attention_offsets)
+                idxs.append(attention_offsets[offset])
+        self.attention_biases = torch.nn.Parameter(torch.zeros(num_heads, len(attention_offsets)))
+        self.register_buffer('attention_bias_idxs', torch.LongTensor(idxs).view(N_, N))
+
+
+    @torch.no_grad()
+    def train(self, mode=True):
+        super().train(mode)
+        if mode and hasattr(self, 'ab'):
+            del self.ab
+        else:
+            self.ab = self.attention_biases[:, self.attention_bias_idxs]
+
+    def forward(self, x):
+        B, N, C = x.shape
+        k, v = self.kv(x).view(B, N, self.num_heads, -1).split([self.key_dim, self.d], dim=3)
+        k = k.permute(0, 2, 1, 3)  # BHNC
+        v = v.permute(0, 2, 1, 3)  # BHNC
+        q = self.q(x).view(B, self.resolution_2, self.num_heads, self.key_dim).permute(0, 2, 1, 3)
+
+        ab = self.attention_biases[:, self.attention_bias_idxs] if self.training else self.ab
+        attn = q @ k.transpose(-2, -1) * self.scale + ab
+        attn = attn.softmax(dim=-1)
+
+        x = (attn @ v).transpose(1, 2).reshape(B, -1, self.dh)
+        x = self.proj(x)
+        return x
+
+
+class Levit(torch.nn.Module):
+    """ Vision Transformer with support for patch or hybrid CNN input stage
+    """
+
+    def __init__(
+            self,
+            img_size=224,
+            patch_size=16,
+            in_chans=3,
+            num_classes=1000,
+            embed_dim=[192],
+            key_dim=[64],
+            depth=[12],
+            num_heads=[3],
+            attn_ratio=[2],
+            mlp_ratio=[2],
+            hybrid_backbone=None,
+            down_ops=[],
+            attn_act_layer=torch.nn.Hardswish,
+            mlp_act_layer=torch.nn.Hardswish,
+            distillation=True,
+            drop_path=0):
+        super().__init__()
+        global FLOPS_COUNTER
+
+        self.num_classes = num_classes
+        self.num_features = embed_dim[-1]
+        self.embed_dim = embed_dim
+        self.distillation = distillation
+
+        self.patch_embed = hybrid_backbone
+
+        self.blocks = []
+        down_ops.append([''])
+        resolution = img_size // patch_size
+        for i, (ed, kd, dpth, nh, ar, mr, do) in enumerate(
+                zip(embed_dim, key_dim, depth, num_heads, attn_ratio, mlp_ratio, down_ops)):
+            for _ in range(dpth):
+                self.blocks.append(
+                    Residual(
+                        Attention(ed, kd, nh, attn_ratio=ar, act_layer=attn_act_layer, resolution=resolution),
+                        drop_path))
+                if mr > 0:
+                    h = int(ed * mr)
+                    self.blocks.append(
+                        Residual(torch.nn.Sequential(
+                            LinearNorm(ed, h, resolution=resolution),
+                            mlp_act_layer(),
+                            LinearNorm(h, ed, bn_weight_init=0, resolution=resolution),
+                        ), drop_path))
+            if do[0] == 'Subsample':
+                # ('Subsample',key_dim, num_heads, attn_ratio, mlp_ratio, stride)
+                resolution_ = (resolution - 1) // do[5] + 1
+                self.blocks.append(
+                    AttentionSubsample(
+                        *embed_dim[i:i + 2], key_dim=do[1], num_heads=do[2],
+                        attn_ratio=do[3], act_layer=attn_act_layer, stride=do[5],
+                        resolution=resolution, resolution_=resolution_))
+                resolution = resolution_
+                if do[4] > 0:  # mlp_ratio
+                    h = int(embed_dim[i + 1] * do[4])
+                    self.blocks.append(
+                        Residual(torch.nn.Sequential(
+                            LinearNorm(embed_dim[i + 1], h, resolution=resolution),
+                            mlp_act_layer(),
+                            LinearNorm(h, embed_dim[i + 1], bn_weight_init=0, resolution=resolution),
+                        ), drop_path))
+        self.blocks = torch.nn.Sequential(*self.blocks)
+
+        # Classifier head
+        self.head = NormLinear(embed_dim[-1], num_classes) if num_classes > 0 else torch.nn.Identity()
+        if distillation:
+            self.head_dist = NormLinear(embed_dim[-1], num_classes) if num_classes > 0 else torch.nn.Identity()
+        else:
+            self.head_dist = None
+
+    @torch.jit.ignore
+    def no_weight_decay(self):
+        return {x for x in self.state_dict().keys() if 'attention_biases' in x}
+
+    def forward(self, x):
+        x = self.patch_embed(x)
+        x = x.flatten(2).transpose(1, 2)
+        x = self.blocks(x)
+        x = x.mean(1)
+        if self.distillation:
+            x = self.head(x), self.head_dist(x)
+            if not self.training:
+                x = (x[0] + x[1]) / 2
+        else:
+            x = self.head(x)
+        return x
+
+
+def model_factory(C, D, X, N, drop_path, weights, num_classes, distillation, pretrained, fuse):
+    embed_dim = [int(x) for x in C.split('_')]
+    num_heads = [int(x) for x in N.split('_')]
+    depth = [int(x) for x in X.split('_')]
+    act = torch.nn.Hardswish
+    model = Levit(
+        patch_size=16,
+        embed_dim=embed_dim,
+        num_heads=num_heads,
+        key_dim=[D] * 3,
+        depth=depth,
+        attn_ratio=[2, 2, 2],
+        mlp_ratio=[2, 2, 2],
+        down_ops=[
+            # ('Subsample',key_dim, num_heads, attn_ratio, mlp_ratio, stride)
+            ['Subsample', D, embed_dim[0] // D, 4, 2, 2],
+            ['Subsample', D, embed_dim[1] // D, 4, 2, 2],
+        ],
+        attn_act_layer=act,
+        mlp_act_layer=act,
+        hybrid_backbone=b16(embed_dim[0], activation=act),
+        num_classes=num_classes,
+        drop_path=drop_path,
+        distillation=distillation
+    )
+    model.default_cfg = _cfg()
+    if pretrained:
+        checkpoint = torch.hub.load_state_dict_from_url(weights, map_location='cpu')
+        model.load_state_dict(checkpoint['model'])
+    #if fuse:
+    #    utils.replace_batchnorm(model)
+
+    return model

timm/models/levitc.py

@@ -0,0 +1,400 @@
+# Copyright (c) 2015-present, Facebook, Inc.
+# All rights reserved.
+
+# Modified from
+# https://github.com/rwightman/pytorch-image-models/blob/master/timm/models/vision_transformer.py
+# Copyright 2020 Ross Wightman, Apache-2.0 License
+import itertools
+
+import torch
+
+from timm.data import IMAGENET_DEFAULT_STD, IMAGENET_DEFAULT_MEAN
+from .vision_transformer import trunc_normal_
+from .registry import register_model
+
+
+def _cfg(url='', **kwargs):
+    return {
+        'url': url,
+        'num_classes': 1000, 'input_size': (3, 224, 224), 'pool_size': None,
+        'crop_pct': .9, 'interpolation': 'bicubic', 'fixed_input_size': True,
+        'mean': IMAGENET_DEFAULT_MEAN, 'std': IMAGENET_DEFAULT_STD,
+        'first_conv': 'patch_embed.proj', 'classifier': 'head',
+        **kwargs
+    }
+
+
+specification = {
+    'levit_c_128s': {
+        'C': '128_256_384', 'D': 16, 'N': '4_6_8', 'X': '2_3_4', 'drop_path': 0,
+        'weights': 'https://dl.fbaipublicfiles.com/LeViT/LeViT-128S-96703c44.pth'},
+    'levit_c_128': {
+        'C': '128_256_384', 'D': 16, 'N': '4_8_12', 'X': '4_4_4', 'drop_path': 0,
+        'weights': 'https://dl.fbaipublicfiles.com/LeViT/LeViT-128-b88c2750.pth'},
+    'levit_c_192': {
+        'C': '192_288_384', 'D': 32, 'N': '3_5_6', 'X': '4_4_4', 'drop_path': 0,
+        'weights': 'https://dl.fbaipublicfiles.com/LeViT/LeViT-192-92712e41.pth'},
+    'levit_c_256': {
+        'C': '256_384_512', 'D': 32, 'N': '4_6_8', 'X': '4_4_4', 'drop_path': 0,
+        'weights': 'https://dl.fbaipublicfiles.com/LeViT/LeViT-256-13b5763e.pth'},
+    'levit_c_384': {
+        'C': '384_512_768', 'D': 32, 'N': '6_9_12', 'X': '4_4_4', 'drop_path': 0.1,
+        'weights': 'https://dl.fbaipublicfiles.com/LeViT/LeViT-384-9bdaf2e2.pth'},
+}
+
+__all__ = ['Levit']
+
+
+@register_model
+def levit_c_128s(num_classes=1000, distillation=True, pretrained=False, fuse=False, **kwargs):
+    return model_factory(**specification['levit_c_128s'], num_classes=num_classes,
+                         distillation=distillation, pretrained=pretrained, fuse=fuse)
+
+
+@register_model
+def levit_c_128(num_classes=1000, distillation=True, pretrained=False, fuse=False, **kwargs):
+    return model_factory(**specification['levit_c_128'], num_classes=num_classes,
+                         distillation=distillation, pretrained=pretrained, fuse=fuse)
+
+
+@register_model
+def levit_c_192(num_classes=1000, distillation=True, pretrained=False, fuse=False, **kwargs):
+    return model_factory(**specification['levit_c_192'], num_classes=num_classes,
+                         distillation=distillation, pretrained=pretrained, fuse=fuse)
+
+
+@register_model
+def levit_c_256(num_classes=1000, distillation=True, pretrained=False, fuse=False, **kwargs):
+    return model_factory(**specification['levit_c_256'], num_classes=num_classes,
+                         distillation=distillation, pretrained=pretrained, fuse=fuse)
+
+
+@register_model
+def levit_c_384(num_classes=1000, distillation=True, pretrained=False, fuse=False, **kwargs):
+    return model_factory(**specification['levit_c_384'], num_classes=num_classes,
+                         distillation=distillation, pretrained=pretrained, fuse=fuse)
+
+
+class ConvNorm(torch.nn.Sequential):
+    def __init__(
+            self, a, b, ks=1, stride=1, pad=0, dilation=1, groups=1, bn_weight_init=1, resolution=-10000):
+        super().__init__()
+        self.add_module('c', torch.nn.Conv2d(a, b, ks, stride, pad, dilation, groups, bias=False))
+        bn = torch.nn.BatchNorm2d(b)
+        torch.nn.init.constant_(bn.weight, bn_weight_init)
+        torch.nn.init.constant_(bn.bias, 0)
+        self.add_module('bn', bn)
+
+    @torch.no_grad()
+    def fuse(self):
+        c, bn = self._modules.values()
+        w = bn.weight / (bn.running_var + bn.eps) ** 0.5
+        w = c.weight * w[:, None, None, None]
+        b = bn.bias - bn.running_mean * bn.weight / \
+            (bn.running_var + bn.eps) ** 0.5
+        m = torch.nn.Conv2d(
+            w.size(1), w.size(0), w.shape[2:], stride=self.c.stride,
+            padding=self.c.padding, dilation=self.c.dilation, groups=self.c.groups)
+        m.weight.data.copy_(w)
+        m.bias.data.copy_(b)
+        return m
+
+
+class NormLinear(torch.nn.Sequential):
+    def __init__(self, a, b, bias=True, std=0.02):
+        super().__init__()
+        self.add_module('bn', torch.nn.BatchNorm1d(a))
+        l = torch.nn.Linear(a, b, bias=bias)
+        trunc_normal_(l.weight, std=std)
+        if bias:
+            torch.nn.init.constant_(l.bias, 0)
+        self.add_module('l', l)
+
+    @torch.no_grad()
+    def fuse(self):
+        bn, l = self._modules.values()
+        w = bn.weight / (bn.running_var + bn.eps) ** 0.5
+        b = bn.bias - self.bn.running_mean * \
+            self.bn.weight / (bn.running_var + bn.eps) ** 0.5
+        w = l.weight * w[None, :]
+        if l.bias is None:
+            b = b @ self.l.weight.T
+        else:
+            b = (l.weight @ b[:, None]).view(-1) + self.l.bias
+        m = torch.nn.Linear(w.size(1), w.size(0))
+        m.weight.data.copy_(w)
+        m.bias.data.copy_(b)
+        return m
+
+
+def b16(n, activation, resolution=224):
+    return torch.nn.Sequential(
+        ConvNorm(3, n // 8, 3, 2, 1, resolution=resolution),
+        activation(),
+        ConvNorm(n // 8, n // 4, 3, 2, 1, resolution=resolution // 2),
+        activation(),
+        ConvNorm(n // 4, n // 2, 3, 2, 1, resolution=resolution // 4),
+        activation(),
+        ConvNorm(n // 2, n, 3, 2, 1, resolution=resolution // 8))
+
+
+class Residual(torch.nn.Module):
+    def __init__(self, m, drop):
+        super().__init__()
+        self.m = m
+        self.drop = drop
+
+    def forward(self, x):
+        if self.training and self.drop > 0:
+            return x + self.m(x) * torch.rand(
+                x.size(0), 1, 1, device=x.device).ge_(self.drop).div(1 - self.drop).detach()
+        else:
+            return x + self.m(x)
+
+
+class Attention(torch.nn.Module):
+    def __init__(self, dim, key_dim, num_heads=8,
+                 attn_ratio=4, act_layer=None, resolution=14):
+        super().__init__()
+        self.num_heads = num_heads
+        self.scale = key_dim ** -0.5
+        self.key_dim = key_dim
+        self.nh_kd = nh_kd = key_dim * num_heads
+        self.d = int(attn_ratio * key_dim)
+        self.dh = int(attn_ratio * key_dim) * num_heads
+        self.attn_ratio = attn_ratio
+        h = self.dh + nh_kd * 2
+        self.qkv = ConvNorm(dim, h, resolution=resolution)
+        self.proj = torch.nn.Sequential(
+            act_layer(),
+            ConvNorm(self.dh, dim, bn_weight_init=0, resolution=resolution))
+
+        points = list(itertools.product(range(resolution), range(resolution)))
+        N = len(points)
+        attention_offsets = {}
+        idxs = []
+        for p1 in points:
+            for p2 in points:
+                offset = (abs(p1[0] - p2[0]), abs(p1[1] - p2[1]))
+                if offset not in attention_offsets:
+                    attention_offsets[offset] = len(attention_offsets)
+                idxs.append(attention_offsets[offset])
+        self.attention_biases = torch.nn.Parameter(torch.zeros(num_heads, len(attention_offsets)))
+        self.register_buffer('attention_bias_idxs', torch.LongTensor(idxs).view(N, N))
+        self.ab = None
+
+    @torch.no_grad()
+    def train(self, mode=True):
+        super().train(mode)
+        if mode and self.ab is not None:
+            self.ab = None
+        else:
+            self.ab = self.attention_biases[:, self.attention_bias_idxs]
+
+    def forward(self, x):  # x (B,C,H,W)
+        B, C, H, W = x.shape
+        q, k, v = self.qkv(x).view(B, self.num_heads, -1, H * W).split([self.key_dim, self.key_dim, self.d], dim=2)
+        ab = self.attention_biases[:, self.attention_bias_idxs] if self.training else self.ab
+        attn = (q.transpose(-2, -1) @ k) * self.scale + ab
+        attn = attn.softmax(dim=-1)
+        x = (v @ attn.transpose(-2, -1)).view(B, -1, H, W)
+        x = self.proj(x)
+        return x
+
+
+class AttentionSubsample(torch.nn.Module):
+    def __init__(
+            self, in_dim, out_dim, key_dim, num_heads=8, attn_ratio=2,
+            act_layer=None, stride=2, resolution=14, resolution_=7):
+        super().__init__()
+        self.num_heads = num_heads
+        self.scale = key_dim ** -0.5
+        self.key_dim = key_dim
+        self.nh_kd = nh_kd = key_dim * num_heads
+        self.d = int(attn_ratio * key_dim)
+        self.dh = int(attn_ratio * key_dim) * self.num_heads
+        self.attn_ratio = attn_ratio
+        self.resolution_ = resolution_
+        self.resolution_2 = resolution_ ** 2
+        h = self.dh + nh_kd
+        self.kv = ConvNorm(in_dim, h, resolution=resolution)
+        self.q = torch.nn.Sequential(
+            torch.nn.AvgPool2d(1, stride, 0),
+            ConvNorm(in_dim, nh_kd, resolution=resolution_))
+        self.proj = torch.nn.Sequential(
+            act_layer(),
+            ConvNorm(self.d * num_heads, out_dim, resolution=resolution_))
+
+        self.stride = stride
+        self.resolution = resolution
+        points = list(itertools.product(range(resolution), range(resolution)))
+        points_ = list(itertools.product(range(resolution_), range(resolution_)))
+        N = len(points)
+        N_ = len(points_)
+        attention_offsets = {}
+        idxs = []
+        for p1 in points_:
+            for p2 in points:
+                size = 1
+                offset = (
+                    abs(p1[0] * stride - p2[0] + (size - 1) / 2),
+                    abs(p1[1] * stride - p2[1] + (size - 1) / 2))
+                if offset not in attention_offsets:
+                    attention_offsets[offset] = len(attention_offsets)
+                idxs.append(attention_offsets[offset])
+        self.attention_biases = torch.nn.Parameter(torch.zeros(num_heads, len(attention_offsets)))
+        self.register_buffer('attention_bias_idxs', torch.LongTensor(idxs).view(N_, N))
+        self.ab = None
+
+    @torch.no_grad()
+    def train(self, mode=True):
+        super().train(mode)
+        if mode and self.ab is not None:
+            self.ab = None
+        else:
+            self.ab = self.attention_biases[:, self.attention_bias_idxs]
+
+    def forward(self, x):
+        B, C, H, W = x.shape
+        k, v = self.kv(x).view(B, self.num_heads, -1, H * W).split([self.key_dim, self.d], dim=2)
+        q = self.q(x).view(B, self.num_heads, self.key_dim, self.resolution_2)
+        ab = self.attention_biases[:, self.attention_bias_idxs] if self.training else self.ab
+        attn = (q.transpose(-2, -1) @ k) * self.scale + ab
+        attn = attn.softmax(dim=-1)
+
+        x = (v @ attn.transpose(-2, -1)).reshape(B, -1, self.resolution_, self.resolution_)
+        x = self.proj(x)
+        return x
+
+
+class Levit(torch.nn.Module):
+    """ Vision Transformer with support for patch or hybrid CNN input stage
+    """
+
+    def __init__(
+            self,
+            img_size=224,
+            patch_size=16,
+            in_chans=3,
+            num_classes=1000,
+            embed_dim=[192],
+            key_dim=[64],
+            depth=[12],
+            num_heads=[3],
+            attn_ratio=[2],
+            mlp_ratio=[2],
+            hybrid_backbone=None,
+            down_ops=[],
+            attn_act_layer=torch.nn.Hardswish,
+            mlp_act_layer=torch.nn.Hardswish,
+            distillation=True,
+            drop_path=0):
+        super().__init__()
+        self.num_classes = num_classes
+        self.num_features = embed_dim[-1]
+        self.embed_dim = embed_dim
+        self.distillation = distillation
+
+        self.patch_embed = hybrid_backbone
+
+        self.blocks = []
+        down_ops.append([''])
+        resolution = img_size // patch_size
+        for i, (ed, kd, dpth, nh, ar, mr, do) in enumerate(
+                zip(embed_dim, key_dim, depth, num_heads, attn_ratio, mlp_ratio, down_ops)):
+            for _ in range(dpth):
+                self.blocks.append(
+                    Residual(
+                        Attention(ed, kd, nh, attn_ratio=ar, act_layer=attn_act_layer, resolution=resolution),
+                        drop_path))
+                if mr > 0:
+                    h = int(ed * mr)
+                    self.blocks.append(
+                        Residual(torch.nn.Sequential(
+                            ConvNorm(ed, h, resolution=resolution),
+                            mlp_act_layer(),
+                            ConvNorm(h, ed, bn_weight_init=0, resolution=resolution),
+                        ), drop_path))
+            if do[0] == 'Subsample':
+                # ('Subsample',key_dim, num_heads, attn_ratio, mlp_ratio, stride)
+                resolution_ = (resolution - 1) // do[5] + 1
+                self.blocks.append(
+                    AttentionSubsample(
+                        *embed_dim[i:i + 2], key_dim=do[1], num_heads=do[2], attn_ratio=do[3],
+                        act_layer=attn_act_layer, stride=do[5],
+                        resolution=resolution, resolution_=resolution_))
+                resolution = resolution_
+                if do[4] > 0:  # mlp_ratio
+                    h = int(embed_dim[i + 1] * do[4])
+                    self.blocks.append(
+                        Residual(torch.nn.Sequential(
+                            ConvNorm(embed_dim[i + 1], h, resolution=resolution),
+                            mlp_act_layer(),
+                            ConvNorm(h, embed_dim[i + 1], bn_weight_init=0, resolution=resolution),
+                        ), drop_path))
+        self.blocks = torch.nn.Sequential(*self.blocks)
+
+        # Classifier head
+        self.head = NormLinear(
+            embed_dim[-1], num_classes) if num_classes > 0 else torch.nn.Identity()
+        if distillation:
+            self.head_dist = NormLinear(
+                embed_dim[-1], num_classes) if num_classes > 0 else torch.nn.Identity()
+
+    @torch.jit.ignore
+    def no_weight_decay(self):
+        return {x for x in self.state_dict().keys() if 'attention_biases' in x}
+
+    def forward(self, x):
+        x = self.patch_embed(x)
+        x = self.blocks(x)
+        x = torch.nn.functional.adaptive_avg_pool2d(x, 1).flatten(1)
+        if self.distillation:
+            x = self.head(x), self.head_dist(x)
+            if not self.training:
+                x = (x[0] + x[1]) / 2
+        else:
+            x = self.head(x)
+        return x
+
+
+def model_factory(C, D, X, N, drop_path, weights, num_classes, distillation, pretrained, fuse):
+    embed_dim = [int(x) for x in C.split('_')]
+    num_heads = [int(x) for x in N.split('_')]
+    depth = [int(x) for x in X.split('_')]
+    act = torch.nn.Hardswish
+    model = Levit(
+        patch_size=16,
+        embed_dim=embed_dim,
+        num_heads=num_heads,
+        key_dim=[D] * 3,
+        depth=depth,
+        attn_ratio=[2, 2, 2],
+        mlp_ratio=[2, 2, 2],
+        down_ops=[
+            # ('Subsample',key_dim, num_heads, attn_ratio, mlp_ratio, stride)
+            ['Subsample', D, embed_dim[0] // D, 4, 2, 2],
+            ['Subsample', D, embed_dim[1] // D, 4, 2, 2],
+        ],
+        attn_act_layer=act,
+        mlp_act_layer=act,
+        hybrid_backbone=b16(embed_dim[0], activation=act),
+        num_classes=num_classes,
+        drop_path=drop_path,
+        distillation=distillation
+    )
+    model.default_cfg = _cfg()
+    if pretrained:
+        checkpoint = torch.hub.load_state_dict_from_url(
+            weights, map_location='cpu')
+        d = checkpoint['model']
+        D = model.state_dict()
+        for k in d.keys():
+            if D[k].shape != d[k].shape:
+                d[k] = d[k][:, :, None, None]
+        model.load_state_dict(d)
+    #if fuse:
+    #    utils.replace_batchnorm(model)
+
+    return model
+

timm/models/visformer.py

@@ -0,0 +1,377 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+from timm.data import IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD
+from .helpers import build_model_with_cfg, overlay_external_default_cfg
+from .layers import to_2tuple, trunc_normal_, DropPath, PatchEmbed
+from .registry import register_model
+
+
+__all__ = ['Visformer']
+
+
+def _cfg(url='', **kwargs):
+    return {
+        'url': url,
+        'num_classes': 1000, 'input_size': (3, 224, 224), 'pool_size': None,
+        'crop_pct': .9, 'interpolation': 'bicubic', 'fixed_input_size': True,
+        'mean': IMAGENET_DEFAULT_MEAN, 'std': IMAGENET_DEFAULT_STD,
+        'first_conv': 'patch_embed.proj', 'classifier': 'head',
+        **kwargs
+    }
+
+
+class LayerNormBHWC(nn.LayerNorm):
+    def __init__(self, dim):
+        super().__init__(dim)
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        return F.layer_norm(
+            x.permute(0, 2, 3, 1), self.normalized_shape, self.weight, self.bias, self.eps).permute(0, 3, 1, 2)
+
+
+class SpatialMlp(nn.Module):
+    def __init__(self, in_features, hidden_features=None, out_features=None,
+                 act_layer=nn.GELU, drop=0., group=8, spatial_conv=False):
+        super().__init__()
+        out_features = out_features or in_features
+        hidden_features = hidden_features or in_features
+        self.in_features = in_features
+        self.out_features = out_features
+        self.spatial_conv = spatial_conv
+        if self.spatial_conv:
+            if group < 2:  # net setting
+                hidden_features = in_features * 5 // 6
+            else:
+                hidden_features = in_features * 2
+        self.hidden_features = hidden_features
+        self.group = group
+        self.drop = nn.Dropout(drop)
+        self.conv1 = nn.Conv2d(in_features, hidden_features, 1, stride=1, padding=0, bias=False)
+        self.act1 = act_layer()
+        if self.spatial_conv:
+            self.conv2 = nn.Conv2d(
+                hidden_features, hidden_features, 3, stride=1, padding=1, groups=self.group, bias=False)
+            self.act2 = act_layer()
+        else:
+            self.conv2 = None
+            self.act2 = None
+        self.conv3 = nn.Conv2d(hidden_features, out_features, 1, stride=1, padding=0, bias=False)
+
+    def forward(self, x):
+        x = self.conv1(x)
+        x = self.act1(x)
+        x = self.drop(x)
+        if self.conv2 is not None:
+            x = self.conv2(x)
+            x = self.act2(x)
+        x = self.conv3(x)
+        x = self.drop(x)
+        return x
+
+
+class Attention(nn.Module):
+    def __init__(self, dim, num_heads=8, head_dim_ratio=1., attn_drop=0., proj_drop=0.):
+        super().__init__()
+        self.dim = dim
+        self.num_heads = num_heads
+        head_dim = round(dim // num_heads * head_dim_ratio)
+        self.head_dim = head_dim
+        self.scale = head_dim ** -0.5
+        self.qkv = nn.Conv2d(dim, head_dim * num_heads * 3, 1, stride=1, padding=0, bias=False)
+        self.attn_drop = nn.Dropout(attn_drop)
+        self.proj = nn.Conv2d(self.head_dim * self.num_heads, dim, 1, stride=1, padding=0, bias=False)
+        self.proj_drop = nn.Dropout(proj_drop)
+
+    def forward(self, x):
+        B, C, H, W = x.shape
+        x = self.qkv(x).reshape(B, 3, self.num_heads, self.head_dim, -1).permute(1, 0, 2, 4, 3)
+        q, k, v = x[0], x[1], x[2]
+
+        attn = (q @ k.transpose(-2, -1)) * self.scale
+        attn = attn.softmax(dim=-1)
+        attn = self.attn_drop(attn)
+        x = attn @ v
+
+        x = x.permute(0, 1, 3, 2).reshape(B, -1, H, W)
+        x = self.proj(x)
+        x = self.proj_drop(x)
+        return x
+
+
+class Block(nn.Module):
+    def __init__(self, dim, num_heads, head_dim_ratio=1., mlp_ratio=4.,
+                 drop=0., attn_drop=0., drop_path=0., act_layer=nn.GELU, norm_layer=LayerNormBHWC,
+                 group=8, attn_disabled=False, spatial_conv=False):
+        super().__init__()
+        self.spatial_conv = spatial_conv
+        self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()
+        if attn_disabled:
+            self.norm1 = None
+            self.attn = None
+        else:
+            self.norm1 = norm_layer(dim)
+            self.attn = Attention(
+                dim, num_heads=num_heads, head_dim_ratio=head_dim_ratio, attn_drop=attn_drop, proj_drop=drop)
+
+        self.norm2 = norm_layer(dim)
+        mlp_hidden_dim = int(dim * mlp_ratio)
+        self.mlp = SpatialMlp(
+            in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop,
+            group=group, spatial_conv=spatial_conv)  # new setting
+
+    def forward(self, x):
+        if self.attn is not None:
+            x = x + self.drop_path(self.attn(self.norm1(x)))
+        x = x + self.drop_path(self.mlp(self.norm2(x)))
+        return x
+
+
+class Visformer(nn.Module):
+    def __init__(self, img_size=224, patch_size=16, in_chans=3, num_classes=1000, init_channels=32, embed_dim=384,
+                 depth=12, num_heads=6, mlp_ratio=4., drop_rate=0., attn_drop_rate=0., drop_path_rate=0.,
+                 norm_layer=LayerNormBHWC, attn_stage='111', pos_embed=True, spatial_conv='111',
+                 vit_stem=False, group=8, pool=True, conv_init=False, embed_norm=None):
+        super().__init__()
+        self.num_classes = num_classes
+        self.num_features = self.embed_dim = embed_dim
+        self.init_channels = init_channels
+        self.img_size = img_size
+        self.vit_stem = vit_stem
+        self.pool = pool
+        self.conv_init = conv_init
+        if isinstance(depth, (list, tuple)):
+            self.stage_num1, self.stage_num2, self.stage_num3 = depth
+            depth = sum(depth)
+        else:
+            self.stage_num1 = self.stage_num3 = depth // 3
+            self.stage_num2 = depth - self.stage_num1 - self.stage_num3
+        self.pos_embed = pos_embed
+        dpr = [x.item() for x in torch.linspace(0, drop_path_rate, depth)]
+
+        # stage 1
+        if self.vit_stem:
+            self.stem = None
+            self.patch_embed1 = PatchEmbed(
+                img_size=img_size, patch_size=patch_size, in_chans=in_chans,
+                embed_dim=embed_dim, norm_layer=embed_norm, flatten=False)
+            img_size //= 16
+        else:
+            if self.init_channels is None:
+                self.stem = None
+                self.patch_embed1 = PatchEmbed(
+                    img_size=img_size, patch_size=patch_size // 2, in_chans=in_chans,
+                    embed_dim=embed_dim // 2, norm_layer=embed_norm, flatten=False)
+                img_size //= 8
+            else:
+                self.stem = nn.Sequential(
+                    nn.Conv2d(3, self.init_channels, 7, stride=2, padding=3, bias=False),
+                    nn.BatchNorm2d(self.init_channels),
+                    nn.ReLU(inplace=True)
+                )
+                img_size //= 2
+                self.patch_embed1 = PatchEmbed(
+                    img_size=img_size, patch_size=patch_size // 4, in_chans=self.init_channels,
+                    embed_dim=embed_dim // 2, norm_layer=embed_norm, flatten=False)
+                img_size //= 4
+
+        if self.pos_embed:
+            if self.vit_stem:
+                self.pos_embed1 = nn.Parameter(torch.zeros(1, embed_dim, img_size, img_size))
+            else:
+                self.pos_embed1 = nn.Parameter(torch.zeros(1, embed_dim//2, img_size, img_size))
+            self.pos_drop = nn.Dropout(p=drop_rate)
+        self.stage1 = nn.ModuleList([
+            Block(
+                dim=embed_dim//2, num_heads=num_heads, head_dim_ratio=0.5, mlp_ratio=mlp_ratio,
+                drop=drop_rate, attn_drop=attn_drop_rate, drop_path=dpr[i], norm_layer=norm_layer,
+                group=group, attn_disabled=(attn_stage[0] == '0'), spatial_conv=(spatial_conv[0] == '1')
+            )
+            for i in range(self.stage_num1)
+        ])
+
+        #stage2
+        if not self.vit_stem:
+            self.patch_embed2 = PatchEmbed(
+                img_size=img_size, patch_size=patch_size // 8, in_chans=embed_dim // 2,
+                embed_dim=embed_dim, norm_layer=embed_norm, flatten=False)
+            img_size //= 2
+            if self.pos_embed:
+                self.pos_embed2 = nn.Parameter(torch.zeros(1, embed_dim, img_size, img_size))
+        self.stage2 = nn.ModuleList([
+            Block(
+                dim=embed_dim, num_heads=num_heads, head_dim_ratio=1.0, mlp_ratio=mlp_ratio,
+                drop=drop_rate, attn_drop=attn_drop_rate, drop_path=dpr[i], norm_layer=norm_layer,
+                group=group, attn_disabled=(attn_stage[1] == '0'), spatial_conv=(spatial_conv[1] == '1')
+            )
+            for i in range(self.stage_num1, self.stage_num1+self.stage_num2)
+        ])
+
+        # stage 3
+        if not self.vit_stem:
+            self.patch_embed3 = PatchEmbed(
+                img_size=img_size, patch_size=patch_size // 8, in_chans=embed_dim,
+                embed_dim=embed_dim * 2, norm_layer=embed_norm, flatten=False)
+            img_size //= 2
+            if self.pos_embed:
+                self.pos_embed3 = nn.Parameter(torch.zeros(1, embed_dim*2, img_size, img_size))
+        self.stage3 = nn.ModuleList([
+            Block(
+                dim=embed_dim*2, num_heads=num_heads, head_dim_ratio=1.0, mlp_ratio=mlp_ratio,
+                drop=drop_rate, attn_drop=attn_drop_rate, drop_path=dpr[i], norm_layer=norm_layer,
+                group=group, attn_disabled=(attn_stage[2] == '0'), spatial_conv=(spatial_conv[2] == '1')
+            )
+            for i in range(self.stage_num1+self.stage_num2, depth)
+        ])
+
+        # head
+        if self.pool:
+            self.global_pooling = nn.AdaptiveAvgPool2d(1)
+        head_dim = embed_dim if self.vit_stem else embed_dim * 2
+        self.norm = norm_layer(head_dim)
+        self.head = nn.Linear(head_dim, num_classes)
+
+        # weights init
+        if self.pos_embed:
+            trunc_normal_(self.pos_embed1, std=0.02)
+            if not self.vit_stem:
+                trunc_normal_(self.pos_embed2, std=0.02)
+                trunc_normal_(self.pos_embed3, std=0.02)
+        self.apply(self._init_weights)
+
+    def _init_weights(self, m):
+        if isinstance(m, nn.Linear):
+            trunc_normal_(m.weight, std=0.02)
+            if 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)
+        elif isinstance(m, nn.BatchNorm2d):
+            nn.init.constant_(m.bias, 0)
+            nn.init.constant_(m.weight, 1.0)
+        elif isinstance(m, nn.Conv2d):
+            if self.conv_init:
+                nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
+            else:
+                trunc_normal_(m.weight, std=0.02)
+            if m.bias is not None:
+                nn.init.constant_(m.bias, 0.)
+
+    def forward(self, x):
+        if self.stem is not None:
+            x = self.stem(x)
+
+        # stage 1
+        x = self.patch_embed1(x)
+        if self.pos_embed:
+            x = x + self.pos_embed1
+            x = self.pos_drop(x)
+        for b in self.stage1:
+            x = b(x)
+
+        # stage 2
+        if not self.vit_stem:
+            x = self.patch_embed2(x)
+            if self.pos_embed:
+                x = x + self.pos_embed2
+                x = self.pos_drop(x)
+        for b in self.stage2:
+            x = b(x)
+
+        # stage3
+        if not self.vit_stem:
+            x = self.patch_embed3(x)
+            if self.pos_embed:
+                x = x + self.pos_embed3
+                x = self.pos_drop(x)
+        for b in self.stage3:
+            x = b(x)
+
+        # head
+        x = self.norm(x)
+        if self.pool:
+            x = self.global_pooling(x)
+        else:
+            x = x[:, :, 0, 0]
+
+        x = self.head(x.view(x.size(0), -1))
+        return x
+
+
+@register_model
+def visformer_tiny(pretrained=False, **kwargs):
+    model = Visformer(
+        img_size=224, init_channels=16, embed_dim=192, depth=(7, 4, 4), num_heads=3, mlp_ratio=4., group=8,
+        attn_stage='011', spatial_conv='100', norm_layer=nn.BatchNorm2d, conv_init=True,
+        embed_norm=nn.BatchNorm2d, **kwargs)
+    return model
+
+
+@register_model
+def visformer_small(pretrained=False, **kwargs):
+    model = Visformer(
+        img_size=224, init_channels=32, embed_dim=384, depth=(7, 4, 4), num_heads=6, mlp_ratio=4., group=8,
+        attn_stage='011', spatial_conv='100', norm_layer=nn.BatchNorm2d, conv_init=True,
+        embed_norm=nn.BatchNorm2d, **kwargs)
+    return model
+
+
+@register_model
+def visformer_net1(pretrained=False, **kwargs):
+    model = Visformer(
+        init_channels=None, embed_dim=384, depth=(0, 12, 0), num_heads=6, mlp_ratio=4., attn_stage='111',
+        spatial_conv='000', vit_stem=True, conv_init=True, **kwargs)
+    return model
+
+
+@register_model
+def visformer_net2(pretrained=False, **kwargs):
+    model = Visformer(
+        init_channels=32, embed_dim=384, depth=(0, 12, 0), num_heads=6, mlp_ratio=4., attn_stage='111',
+        spatial_conv='000', vit_stem=False, conv_init=True, **kwargs)
+    return model
+
+
+@register_model
+def visformer_net3(pretrained=False, **kwargs):
+    model = Visformer(
+        init_channels=32, embed_dim=384, depth=12, num_heads=6, mlp_ratio=4., attn_stage='111',
+        spatial_conv='000', vit_stem=False, conv_init=True, **kwargs)
+    return model
+
+
+@register_model
+def visformer_net4(pretrained=False, **kwargs):
+    model = Visformer(init_channels=32, embed_dim=384, depth=12, num_heads=6, mlp_ratio=4., attn_stage='111',
+                      spatial_conv='000', vit_stem=False, conv_init=True, **kwargs)
+    return model
+
+
+@register_model
+def visformer_net5(pretrained=False, **kwargs):
+    model = Visformer(
+        init_channels=32, embed_dim=384, depth=12, num_heads=6, mlp_ratio=4., group=1, attn_stage='111',
+        spatial_conv='111', vit_stem=False, conv_init=True, **kwargs)
+    return model
+
+
+@register_model
+def visformer_net6(pretrained=False, **kwargs):
+    model = Visformer(
+        init_channels=32, embed_dim=384, depth=12, num_heads=6, mlp_ratio=4., group=1, attn_stage='111',
+        pos_embed=False, spatial_conv='111', conv_init=True, **kwargs)
+    return model
+
+
+@register_model
+def visformer_net7(pretrained=False, **kwargs):
+    model = Visformer(
+        init_channels=32, embed_dim=384, depth=(6, 7, 7), num_heads=6, group=1, attn_stage='000',
+        pos_embed=False, spatial_conv='111', conv_init=True, **kwargs)
+    return model
+
+
+
+