File size: 24,440 Bytes
bcfa144
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
import os.path

import torch
import torch.nn as nn
from functools import partial

from timm.models.vision_transformer import Mlp, PatchEmbed , _cfg

from timm.models.layers import DropPath, to_2tuple, trunc_normal_
from timm.models.registry import register_model
# from xformers.ops import memory_efficient_attention

class Attention(nn.Module):
    # taken from https://github.com/rwightman/pytorch-image-models/blob/master/timm/models/vision_transformer.py
    def __init__(self, dim, num_heads=8, qkv_bias=False, qk_scale=None, attn_drop=0., proj_drop=0.):
        super().__init__()
        self.num_heads = num_heads
        head_dim = dim // num_heads
        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)

    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]


        # x = memory_efficient_attention(q, k, v).transpose(1, 2).reshape(B, N, C)

        q = q * self.scale
        attn = (q @ k.transpose(-2, -1))
        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):
    # taken from https://github.com/rwightman/pytorch-image-models/blob/master/timm/models/vision_transformer.py
    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,Attention_block = Attention,Mlp_block=Mlp
                 ,init_values=1e-4):
        super().__init__()
        self.norm1 = norm_layer(dim)
        self.attn = Attention_block(
            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_block(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 Layer_scale_init_Block(nn.Module):
    # taken from https://github.com/rwightman/pytorch-image-models/blob/master/timm/models/vision_transformer.py
    # with slight modifications
    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,Attention_block = Attention,Mlp_block=Mlp
                 ,init_values=1e-4):
        super().__init__()
        self.norm1 = norm_layer(dim)
        self.attn = Attention_block(
            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_block(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop)
        self.gamma_1 = nn.Parameter(init_values * torch.ones((dim)),requires_grad=True)
        self.gamma_2 = nn.Parameter(init_values * torch.ones((dim)),requires_grad=True)

    def forward(self, x):
        x = x + self.drop_path(self.gamma_1 * self.attn(self.norm1(x)))
        x = x + self.drop_path(self.gamma_2 * self.mlp(self.norm2(x)))
        return x

class Layer_scale_init_Block_paralx2(nn.Module):
    # taken from https://github.com/rwightman/pytorch-image-models/blob/master/timm/models/vision_transformer.py
    # with slight modifications
    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,Attention_block = Attention,Mlp_block=Mlp
                 ,init_values=1e-4):
        super().__init__()
        self.norm1 = norm_layer(dim)
        self.norm11 = norm_layer(dim)
        self.attn = Attention_block(
            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.attn1 = Attention_block(
            dim, num_heads=num_heads, qkv_bias=qkv_bias, qk_scale=qk_scale, attn_drop=attn_drop, proj_drop=drop)
        self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()
        self.norm2 = norm_layer(dim)
        self.norm21 = norm_layer(dim)
        mlp_hidden_dim = int(dim * mlp_ratio)
        self.mlp = Mlp_block(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop)
        self.mlp1 = Mlp_block(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop)
        self.gamma_1 = nn.Parameter(init_values * torch.ones((dim)),requires_grad=True)
        self.gamma_1_1 = nn.Parameter(init_values * torch.ones((dim)),requires_grad=True)
        self.gamma_2 = nn.Parameter(init_values * torch.ones((dim)),requires_grad=True)
        self.gamma_2_1 = nn.Parameter(init_values * torch.ones((dim)),requires_grad=True)
        
    def forward(self, x):
        x = x + self.drop_path(self.gamma_1*self.attn(self.norm1(x))) + self.drop_path(self.gamma_1_1 * self.attn1(self.norm11(x)))
        x = x + self.drop_path(self.gamma_2 * self.mlp(self.norm2(x))) + self.drop_path(self.gamma_2_1 * self.mlp1(self.norm21(x)))
        return x
        
class Block_paralx2(nn.Module):
    # taken from https://github.com/rwightman/pytorch-image-models/blob/master/timm/models/vision_transformer.py
    # with slight modifications
    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,Attention_block = Attention,Mlp_block=Mlp
                 ,init_values=1e-4):
        super().__init__()
        self.norm1 = norm_layer(dim)
        self.norm11 = norm_layer(dim)
        self.attn = Attention_block(
            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.attn1 = Attention_block(
            dim, num_heads=num_heads, qkv_bias=qkv_bias, qk_scale=qk_scale, attn_drop=attn_drop, proj_drop=drop)
        self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()
        self.norm2 = norm_layer(dim)
        self.norm21 = norm_layer(dim)
        mlp_hidden_dim = int(dim * mlp_ratio)
        self.mlp = Mlp_block(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop)
        self.mlp1 = Mlp_block(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))) + self.drop_path(self.attn1(self.norm11(x)))
        x = x + self.drop_path(self.mlp(self.norm2(x))) + self.drop_path(self.mlp1(self.norm21(x)))
        return x
        
        
class hMLP_stem(nn.Module):
    """ hMLP_stem: https://arxiv.org/pdf/2203.09795.pdf
    taken from https://github.com/rwightman/pytorch-image-models/blob/master/timm/models/vision_transformer.py
    with slight modifications
    """
    def __init__(self, img_size=224,  patch_size=16, in_chans=3, embed_dim=768,norm_layer=nn.SyncBatchNorm):
        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 = torch.nn.Sequential(*[nn.Conv2d(in_chans, embed_dim//4, kernel_size=4, stride=4),
                                          norm_layer(embed_dim//4),
                                          nn.GELU(),
                                          nn.Conv2d(embed_dim//4, embed_dim//4, kernel_size=2, stride=2),
                                          norm_layer(embed_dim//4),
                                          nn.GELU(),
                                          nn.Conv2d(embed_dim//4, embed_dim, kernel_size=2, stride=2),
                                          norm_layer(embed_dim),
                                         ])
        

    def forward(self, x):
        B, C, H, W = x.shape
        x = self.proj(x).flatten(2).transpose(1, 2)
        return x
    
class vit_models(nn.Module):
    """ Vision Transformer with LayerScale (https://arxiv.org/abs/2103.17239) support
    taken from https://github.com/rwightman/pytorch-image-models/blob/master/timm/models/vision_transformer.py
    with slight modifications
    """
    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=False, qk_scale=None, drop_rate=0., attn_drop_rate=0.,
                 drop_path_rate=0., norm_layer=nn.LayerNorm, global_pool=None,
                 block_layers = Block,
                 Patch_layer=PatchEmbed,act_layer=nn.GELU,
                 Attention_block = Attention, Mlp_block=Mlp,
                dpr_constant=True,init_scale=1e-4,
                mlp_ratio_clstk = 4.0):
        super().__init__()
        
        self.dropout_rate = drop_rate
        self.depth = depth

        self.num_classes = num_classes
        self.num_features = self.embed_dim = embed_dim

        self.patch_embed = Patch_layer(
                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, embed_dim))

        dpr = [drop_path_rate for i in range(depth)]
        self.blocks = nn.ModuleList([
            block_layers(
                dim=embed_dim, num_heads=num_heads, mlp_ratio=mlp_ratio, qkv_bias=qkv_bias, qk_scale=qk_scale,
                drop=0.0, attn_drop=attn_drop_rate, drop_path=dpr[i], norm_layer=norm_layer,
                act_layer=act_layer,Attention_block=Attention_block,Mlp_block=Mlp_block,init_values=init_scale)
            for i in range(depth)])

        self.norm = norm_layer(embed_dim)

        self.feature_info = [dict(num_chs=embed_dim, reduction=0, module='head')]
        self.head = nn.Linear(embed_dim, 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 get_num_layers(self):
        return len(self.blocks)
    
    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 extract_block_features(self, x):
        B = x.shape[0]
        x = self.patch_embed(x)

        cls_tokens = self.cls_token.expand(B, -1, -1)

        x = x + self.pos_embed

        x = torch.cat((cls_tokens, x), dim=1)

        outs = {}

        for i, blk in enumerate(self.blocks):
            x = blk(x)
            outs[i] = x.detach()
        return outs

    def selective_forward(self, x, begin, end):
        for i, blk in enumerate(self.blocks):
            if i < begin:
                continue
            if i > end:
                break
            x = blk(x)
        return x

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

        cls_tokens = self.cls_token.expand(B, -1, -1)
        x = x + self.pos_embed
        x = torch.cat((cls_tokens, x), dim=1)

        for i, blk in enumerate(self.blocks):
            x = blk(x)
            if i == blk_id:
                break

        return x

    def forward_from(self, x, blk_id):
        for i, blk in enumerate(self.blocks):
            if i < blk_id:
                continue
            x = blk(x)

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

        return x

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

        cls_tokens = self.cls_token.expand(B, -1, -1)
        
        x = x + self.pos_embed
        
        x = torch.cat((cls_tokens, x), dim=1)
        return x

    
    def forward_norm_head(self, x):
        x = self.norm(x)
        x = self.head(x[:, 0])
        return x


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

        cls_tokens = self.cls_token.expand(B, -1, -1)
        
        x = x + self.pos_embed
        
        x = torch.cat((cls_tokens, x), dim=1)
            
        for i , blk in enumerate(self.blocks):
            x = blk(x)
            
        x = self.norm(x)
        return x[:, 0]

    def forward(self, x):

        x = self.forward_features(x)
        
        if self.dropout_rate:
            x = F.dropout(x, p=float(self.dropout_rate), training=self.training)
        x = self.head(x)
        
        return x

# DeiT III: Revenge of the ViT (https://arxiv.org/abs/2204.07118)

@register_model
def deit_tiny_patch16_LS(pretrained=False, img_size=224, pretrained_21k = False,  pretrained_cfg_overlay=None, **kwargs):
    model = vit_models(
        img_size = img_size, patch_size=16, embed_dim=192, depth=12, num_heads=3, mlp_ratio=4, qkv_bias=True,
        norm_layer=partial(nn.LayerNorm, eps=1e-6),block_layers=Layer_scale_init_Block, **kwargs)
    
    return model
    
    
@register_model
def deit_small_patch16_LS(pretrained=False, img_size=224, pretrained_21k = False, pretrained_cfg=None, pretrained_deit=None, pretrained_cfg_overlay=None, **kwargs):
    model = vit_models(
        img_size = img_size, patch_size=16, embed_dim=384, depth=12, num_heads=6, mlp_ratio=4, qkv_bias=True,
        norm_layer=partial(nn.LayerNorm, eps=1e-6),block_layers=Layer_scale_init_Block, **kwargs)
    model.default_cfg = _cfg()
    if pretrained:
        # name = 'https://dl.fbaipublicfiles.com/deit/deit_3_small_'+str(img_size)+'_'
        # if pretrained_21k:
        #     name+='21k.pth'
        # else:
        #     name+='1k.pth'
            
        # checkpoint = torch.hub.load_state_dict_from_url(
        #     url=name,
        #     map_location="cpu", check_hash=True
        # )
        checkpoint = torch.load(os.path.join(pretrained_deit, 'deit_3_small_224_21k.pth'))
        model.load_state_dict(checkpoint["model"])

    return model

@register_model
def deit_medium_patch16_LS(pretrained=False, img_size=224, pretrained_21k = False, **kwargs):
    model = vit_models(
        patch_size=16, embed_dim=512, depth=12, num_heads=8, mlp_ratio=4, qkv_bias=True,
        norm_layer=partial(nn.LayerNorm, eps=1e-6),block_layers = Layer_scale_init_Block, **kwargs)
    model.default_cfg = _cfg()
    if pretrained:
        name = 'https://dl.fbaipublicfiles.com/deit/deit_3_medium_'+str(img_size)+'_'
        if pretrained_21k:
            name+='21k.pth'
        else:
            name+='1k.pth'
            
        checkpoint = torch.hub.load_state_dict_from_url(
            url=name,
            map_location="cpu", check_hash=True
        )
        model.load_state_dict(checkpoint["model"])
    return model 

@register_model
def deit_base_patch16_LS(pretrained=False, pretrained_cfg=None, img_size=224, pretrained_21k = False,  pretrained_deit=None, pretrained_cfg_overlay=None, **kwargs):
    model = vit_models(
        img_size = img_size, patch_size=16, embed_dim=768, depth=12, num_heads=12, mlp_ratio=4, qkv_bias=True,
        norm_layer=partial(nn.LayerNorm, eps=1e-6),block_layers=Layer_scale_init_Block, **kwargs)
    if pretrained:
        # name = 'https://dl.fbaipublicfiles.com/deit/deit_3_small_'+str(img_size)+'_'
        # if pretrained_21k:
        #     name+='21k.pth'
        # else:
        #     name+='1k.pth'

        # checkpoint = torch.hub.load_state_dict_from_url(
        #     url=name,
        #     map_location="cpu", check_hash=True
        # )
        checkpoint = torch.load(os.path.join(pretrained_deit, 'deit_3_base_224_21k.pth'))
        model.load_state_dict(checkpoint["model"])
    return model
    
@register_model
def deit_large_patch16_LS(pretrained=False, img_size=224, pretrained_21k = False,  pretrained_cfg=None, pretrained_deit=None, pretrained_cfg_overlay=None, **kwargs):
    model = vit_models(
        img_size = img_size, patch_size=16, embed_dim=1024, depth=24, num_heads=16, mlp_ratio=4, qkv_bias=True,
        norm_layer=partial(nn.LayerNorm, eps=1e-6),block_layers=Layer_scale_init_Block, **kwargs)
    if pretrained:
        # name = 'https://dl.fbaipublicfiles.com/deit/deit_3_large_'+str(img_size)+'_'
        # if pretrained_21k:
        #     name+='21k.pth'
        # else:
        #     name+='1k.pth'
        #
        # checkpoint = torch.hub.load_state_dict_from_url(
        #     url=name,
        #     map_location="cpu", check_hash=True
        # )
        checkpoint = torch.load(os.path.join(pretrained_deit, 'deit_3_large_224_21k.pth'))
        model.load_state_dict(checkpoint["model"])
    return model
    
@register_model
def deit_huge_patch14_LS(pretrained=False, img_size=224, pretrained_21k = False,  **kwargs):
    model = vit_models(
        img_size = img_size, patch_size=14, embed_dim=1280, depth=32, num_heads=16, mlp_ratio=4, qkv_bias=True,
        norm_layer=partial(nn.LayerNorm, eps=1e-6),block_layers = Layer_scale_init_Block, **kwargs)
    if pretrained:
        name = 'https://dl.fbaipublicfiles.com/deit/deit_3_huge_'+str(img_size)+'_'
        if pretrained_21k:
            name+='21k_v1.pth'
        else:
            name+='1k_v1.pth'
            
        checkpoint = torch.hub.load_state_dict_from_url(
            url=name,
            map_location="cpu", check_hash=True
        )
        model.load_state_dict(checkpoint["model"])
    return model
    
@register_model
def deit_huge_patch14_52_LS(pretrained=False, img_size=224, pretrained_21k = False,  **kwargs):
    model = vit_models(
        img_size = img_size, patch_size=14, embed_dim=1280, depth=52, num_heads=16, mlp_ratio=4, qkv_bias=True,
        norm_layer=partial(nn.LayerNorm, eps=1e-6),block_layers = Layer_scale_init_Block, **kwargs)

    return model
    
@register_model
def deit_huge_patch14_26x2_LS(pretrained=False, img_size=224, pretrained_21k = False,  **kwargs):
    model = vit_models(
        img_size = img_size, patch_size=14, embed_dim=1280, depth=26, num_heads=16, mlp_ratio=4, qkv_bias=True,
        norm_layer=partial(nn.LayerNorm, eps=1e-6),block_layers = Layer_scale_init_Block_paralx2, **kwargs)

    return model
    
@register_model
def deit_Giant_48x2_patch14_LS(pretrained=False, img_size=224, pretrained_21k = False,  **kwargs):
    model = vit_models(
        img_size = img_size, patch_size=14, embed_dim=1664, depth=48, num_heads=16, mlp_ratio=4, qkv_bias=True,
        norm_layer=partial(nn.LayerNorm, eps=1e-6),block_layers = Block_paral_LS, **kwargs)

    return model

@register_model
def deit_giant_40x2_patch14_LS(pretrained=False, img_size=224, pretrained_21k = False,  **kwargs):
    model = vit_models(
        img_size = img_size, patch_size=14, embed_dim=1408, depth=40, num_heads=16, mlp_ratio=4, qkv_bias=True,
        norm_layer=partial(nn.LayerNorm, eps=1e-6),block_layers = Block_paral_LS, **kwargs)
    return model

@register_model
def deit_Giant_48_patch14_LS(pretrained=False, img_size=224, pretrained_21k = False,  **kwargs):
    model = vit_models(
        img_size = img_size, patch_size=14, embed_dim=1664, depth=48, num_heads=16, mlp_ratio=4, qkv_bias=True,
        norm_layer=partial(nn.LayerNorm, eps=1e-6),block_layers = Layer_scale_init_Block, **kwargs)
    return model

@register_model
def deit_giant_40_patch14_LS(pretrained=False, img_size=224, pretrained_21k = False,  **kwargs):
    model = vit_models(
        img_size = img_size, patch_size=14, embed_dim=1408, depth=40, num_heads=16, mlp_ratio=4, qkv_bias=True,
        norm_layer=partial(nn.LayerNorm, eps=1e-6),block_layers = Layer_scale_init_Block, **kwargs)
    #model.default_cfg = _cfg()

    return model

# Models from Three things everyone should know about Vision Transformers (https://arxiv.org/pdf/2203.09795.pdf)

@register_model
def deit_small_patch16_36_LS(pretrained=False, img_size=224, pretrained_21k = False,  **kwargs):
    model = vit_models(
        img_size = img_size, patch_size=16, embed_dim=384, depth=36, num_heads=6, mlp_ratio=4, qkv_bias=True,
        norm_layer=partial(nn.LayerNorm, eps=1e-6),block_layers=Layer_scale_init_Block, **kwargs)

    return model
    
@register_model
def deit_small_patch16_36(pretrained=False, img_size=224, pretrained_21k = False,  **kwargs):
    model = vit_models(
        img_size = img_size, patch_size=16, embed_dim=384, depth=36, num_heads=6, mlp_ratio=4, qkv_bias=True,
        norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)

    return model
    
@register_model
def deit_small_patch16_18x2_LS(pretrained=False, img_size=224, pretrained_21k = False,  **kwargs):
    model = vit_models(
        img_size = img_size, patch_size=16, embed_dim=384, depth=18, num_heads=6, mlp_ratio=4, qkv_bias=True,
        norm_layer=partial(nn.LayerNorm, eps=1e-6),block_layers=Layer_scale_init_Block_paralx2, **kwargs)

    return model
    
@register_model
def deit_small_patch16_18x2(pretrained=False, img_size=224, pretrained_21k = False,  **kwargs):
    model = vit_models(
        img_size = img_size, patch_size=16, embed_dim=384, depth=18, num_heads=6, mlp_ratio=4, qkv_bias=True,
        norm_layer=partial(nn.LayerNorm, eps=1e-6),block_layers=Block_paralx2, **kwargs)

    return model
    
  
@register_model
def deit_base_patch16_18x2_LS(pretrained=False, img_size=224, pretrained_21k = False,  **kwargs):
    model = vit_models(
        img_size = img_size, patch_size=16, embed_dim=768, depth=18, num_heads=12, mlp_ratio=4, qkv_bias=True,
        norm_layer=partial(nn.LayerNorm, eps=1e-6),block_layers=Layer_scale_init_Block_paralx2, **kwargs)

    return model


@register_model
def deit_base_patch16_18x2(pretrained=False, img_size=224, pretrained_21k = False,  **kwargs):
    model = vit_models(
        img_size = img_size, patch_size=16, embed_dim=768, depth=18, num_heads=12, mlp_ratio=4, qkv_bias=True,
        norm_layer=partial(nn.LayerNorm, eps=1e-6),block_layers=Block_paralx2, **kwargs)

    return model
    

@register_model
def deit_base_patch16_36x1_LS(pretrained=False, img_size=224, pretrained_21k = False,  **kwargs):
    model = vit_models(
        img_size = img_size, patch_size=16, embed_dim=768, depth=36, num_heads=12, mlp_ratio=4, qkv_bias=True,
        norm_layer=partial(nn.LayerNorm, eps=1e-6),block_layers=Layer_scale_init_Block, **kwargs)

    return model

@register_model
def deit_base_patch16_36x1(pretrained=False, img_size=224, pretrained_21k = False,  **kwargs):
    model = vit_models(
        img_size = img_size, patch_size=16, embed_dim=768, depth=36, num_heads=12, mlp_ratio=4, qkv_bias=True,
        norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)

    return model