""" Vision Transformer (ViT) in PyTorch
A PyTorch implement of Vision Transformers as described in:
'An Image Is Worth 16 x 16 Words: Transformers for Image Recognition at Scale'
- https://arxiv.org/abs/2010.11929
`How to train your ViT? Data, Augmentation, and Regularization in Vision Transformers`
- https://arxiv.org/abs/2106.10270
The official jax code is released and available at https://github.com/google-research/vision_transformer
DeiT model defs and weights from https://github.com/facebookresearch/deit,
paper `DeiT: Data-efficient Image Transformers` - https://arxiv.org/abs/2012.12877
Acknowledgments:
* The paper authors for releasing code and weights, thanks!
* I fixed my class token impl based on Phil Wang's https://github.com/lucidrains/vit-pytorch ... check it out
for some einops/einsum fun
* Simple transformer style inspired by Andrej Karpathy's https://github.com/karpathy/minGPT
* Bert reference code checks against Huggingface Transformers and Tensorflow Bert
Hacked together by / Copyright 2020, Ross Wightman
"""
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, IMAGENET_INCEPTION_MEAN, IMAGENET_INCEPTION_STD
from timm.models.helpers import build_model_with_cfg, named_apply, adapt_input_conv, resolve_pretrained_cfg
from timm.models.layers import PatchEmbed, Mlp, DropPath, trunc_normal_, lecun_normal_
_logger = logging.getLogger(__name__)
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_INCEPTION_MEAN, 'std': IMAGENET_INCEPTION_STD,
'first_conv': 'patch_embed.proj', 'classifier': 'head',
**kwargs
}
default_cfgs = {
# patch models (weights from official Google JAX impl)
'vit_tiny_patch16_224': _cfg(
url='https://storage.googleapis.com/vit_models/augreg/'
'Ti_16-i21k-300ep-lr_0.001-aug_none-wd_0.03-do_0.0-sd_0.0--imagenet2012-steps_20k-lr_0.03-res_224.npz'),
'vit_tiny_patch16_384': _cfg(
url='https://storage.googleapis.com/vit_models/augreg/'
'Ti_16-i21k-300ep-lr_0.001-aug_none-wd_0.03-do_0.0-sd_0.0--imagenet2012-steps_20k-lr_0.03-res_384.npz',
input_size=(3, 384, 384), crop_pct=1.0),
'vit_small_patch32_224': _cfg(
url='https://storage.googleapis.com/vit_models/augreg/'
'S_32-i21k-300ep-lr_0.001-aug_light1-wd_0.03-do_0.0-sd_0.0--imagenet2012-steps_20k-lr_0.03-res_224.npz'),
'vit_small_patch32_384': _cfg(
url='https://storage.googleapis.com/vit_models/augreg/'
'S_32-i21k-300ep-lr_0.001-aug_light1-wd_0.03-do_0.0-sd_0.0--imagenet2012-steps_20k-lr_0.03-res_384.npz',
input_size=(3, 384, 384), crop_pct=1.0),
'vit_small_patch16_224': _cfg(
url='https://storage.googleapis.com/vit_models/augreg/'
'S_16-i21k-300ep-lr_0.001-aug_light1-wd_0.03-do_0.0-sd_0.0--imagenet2012-steps_20k-lr_0.03-res_224.npz'),
'vit_small_patch16_384': _cfg(
url='https://storage.googleapis.com/vit_models/augreg/'
'S_16-i21k-300ep-lr_0.001-aug_light1-wd_0.03-do_0.0-sd_0.0--imagenet2012-steps_20k-lr_0.03-res_384.npz',
input_size=(3, 384, 384), crop_pct=1.0),
'vit_base_patch32_224': _cfg(
url='https://storage.googleapis.com/vit_models/augreg/'
'B_32-i21k-300ep-lr_0.001-aug_medium1-wd_0.03-do_0.0-sd_0.0--imagenet2012-steps_20k-lr_0.03-res_224.npz'),
'vit_base_patch32_384': _cfg(
url='https://storage.googleapis.com/vit_models/augreg/'
'B_32-i21k-300ep-lr_0.001-aug_light1-wd_0.1-do_0.0-sd_0.0--imagenet2012-steps_20k-lr_0.03-res_384.npz',
input_size=(3, 384, 384), crop_pct=1.0),
'vit_base_patch16_224': _cfg(
url='https://storage.googleapis.com/vit_models/augreg/'
'B_16-i21k-300ep-lr_0.001-aug_medium1-wd_0.1-do_0.0-sd_0.0--imagenet2012-steps_20k-lr_0.01-res_224.npz'),
'vit_base_patch16_384': _cfg(
url='https://storage.googleapis.com/vit_models/augreg/'
'B_16-i21k-300ep-lr_0.001-aug_medium1-wd_0.1-do_0.0-sd_0.0--imagenet2012-steps_20k-lr_0.01-res_384.npz',
input_size=(3, 384, 384), crop_pct=1.0),
'vit_base_patch8_224': _cfg(
url='https://storage.googleapis.com/vit_models/augreg/'
'B_8-i21k-300ep-lr_0.001-aug_medium1-wd_0.1-do_0.0-sd_0.0--imagenet2012-steps_20k-lr_0.01-res_224.npz'),
'vit_large_patch32_224': _cfg(
url='', # no official model weights for this combo, only for in21k
),
'vit_large_patch32_384': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-vitjx/jx_vit_large_p32_384-9b920ba8.pth',
input_size=(3, 384, 384), crop_pct=1.0),
'vit_large_patch16_224': _cfg(
url='https://storage.googleapis.com/vit_models/augreg/'
'L_16-i21k-300ep-lr_0.001-aug_medium1-wd_0.1-do_0.1-sd_0.1--imagenet2012-steps_20k-lr_0.01-res_224.npz'),
'vit_large_patch16_384': _cfg(
url='https://storage.googleapis.com/vit_models/augreg/'
'L_16-i21k-300ep-lr_0.001-aug_medium1-wd_0.1-do_0.1-sd_0.1--imagenet2012-steps_20k-lr_0.01-res_384.npz',
input_size=(3, 384, 384), crop_pct=1.0),
'vit_huge_patch14_224': _cfg(url=''),
'vit_giant_patch14_224': _cfg(url=''),
'vit_gigantic_patch14_224': _cfg(url=''),
# patch models, imagenet21k (weights from official Google JAX impl)
'vit_tiny_patch16_224_in21k': _cfg(
url='https://storage.googleapis.com/vit_models/augreg/Ti_16-i21k-300ep-lr_0.001-aug_none-wd_0.03-do_0.0-sd_0.0.npz',
num_classes=21843),
'vit_small_patch32_224_in21k': _cfg(
url='https://storage.googleapis.com/vit_models/augreg/S_32-i21k-300ep-lr_0.001-aug_light1-wd_0.03-do_0.0-sd_0.0.npz',
num_classes=21843),
'vit_small_patch16_224_in21k': _cfg(
url='https://storage.googleapis.com/vit_models/augreg/S_16-i21k-300ep-lr_0.001-aug_light1-wd_0.03-do_0.0-sd_0.0.npz',
num_classes=21843),
'vit_base_patch32_224_in21k': _cfg(
url='https://storage.googleapis.com/vit_models/augreg/B_32-i21k-300ep-lr_0.001-aug_medium1-wd_0.03-do_0.0-sd_0.0.npz',
num_classes=21843),
'vit_base_patch16_224_in21k': _cfg(
url='https://storage.googleapis.com/vit_models/augreg/B_16-i21k-300ep-lr_0.001-aug_medium1-wd_0.1-do_0.0-sd_0.0.npz',
# url='./B_16-i21k-300ep-lr_0.001-aug_medium1-wd_0.1-do_0.0-sd_0.0.npz',
num_classes=21843),
'vit_base_patch8_224_in21k': _cfg(
url='https://storage.googleapis.com/vit_models/augreg/B_8-i21k-300ep-lr_0.001-aug_medium1-wd_0.1-do_0.0-sd_0.0.npz',
num_classes=21843),
'vit_large_patch32_224_in21k': _cfg(
url='https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-vitjx/jx_vit_large_patch32_224_in21k-9046d2e7.pth',
num_classes=21843),
'vit_large_patch16_224_in21k': _cfg(
url='https://storage.googleapis.com/vit_models/augreg/L_16-i21k-300ep-lr_0.001-aug_medium1-wd_0.1-do_0.1-sd_0.1.npz',
num_classes=21843),
'vit_huge_patch14_224_in21k': _cfg(
url='https://storage.googleapis.com/vit_models/imagenet21k/ViT-H_14.npz',
hf_hub='timm/vit_huge_patch14_224_in21k',
num_classes=21843),
# SAM trained models (https://arxiv.org/abs/2106.01548)
'vit_base_patch32_sam_224': _cfg(
url='https://storage.googleapis.com/vit_models/sam/ViT-B_32.npz'),
'vit_base_patch16_sam_224': _cfg(
url='https://storage.googleapis.com/vit_models/sam/ViT-B_16.npz'),
# deit models (FB weights)
'deit_tiny_patch16_224': _cfg(
url='https://dl.fbaipublicfiles.com/deit/deit_tiny_patch16_224-a1311bcf.pth',
mean=IMAGENET_DEFAULT_MEAN, std=IMAGENET_DEFAULT_STD),
'deit_small_patch16_224': _cfg(
url='https://dl.fbaipublicfiles.com/deit/deit_small_patch16_224-cd65a155.pth',
mean=IMAGENET_DEFAULT_MEAN, std=IMAGENET_DEFAULT_STD),
'deit_base_patch16_224': _cfg(
url='https://dl.fbaipublicfiles.com/deit/deit_base_patch16_224-b5f2ef4d.pth',
mean=IMAGENET_DEFAULT_MEAN, std=IMAGENET_DEFAULT_STD),
'deit_base_patch16_384': _cfg(
url='https://dl.fbaipublicfiles.com/deit/deit_base_patch16_384-8de9b5d1.pth',
mean=IMAGENET_DEFAULT_MEAN, std=IMAGENET_DEFAULT_STD, input_size=(3, 384, 384), crop_pct=1.0),
'deit_tiny_distilled_patch16_224': _cfg(
url='https://dl.fbaipublicfiles.com/deit/deit_tiny_distilled_patch16_224-b40b3cf7.pth',
mean=IMAGENET_DEFAULT_MEAN, std=IMAGENET_DEFAULT_STD, classifier=('head', 'head_dist')),
'deit_small_distilled_patch16_224': _cfg(
url='https://dl.fbaipublicfiles.com/deit/deit_small_distilled_patch16_224-649709d9.pth',
mean=IMAGENET_DEFAULT_MEAN, std=IMAGENET_DEFAULT_STD, classifier=('head', 'head_dist')),
'deit_base_distilled_patch16_224': _cfg(
url='https://dl.fbaipublicfiles.com/deit/deit_base_distilled_patch16_224-df68dfff.pth',
mean=IMAGENET_DEFAULT_MEAN, std=IMAGENET_DEFAULT_STD, classifier=('head', 'head_dist')),
'deit_base_distilled_patch16_384': _cfg(
url='https://dl.fbaipublicfiles.com/deit/deit_base_distilled_patch16_384-d0272ac0.pth',
mean=IMAGENET_DEFAULT_MEAN, std=IMAGENET_DEFAULT_STD, input_size=(3, 384, 384), crop_pct=1.0,
classifier=('head', 'head_dist')),
# ViT ImageNet-21K-P pretraining by MILL
'vit_base_patch16_224_miil_in21k': _cfg(
url='https://miil-public-eu.oss-eu-central-1.aliyuncs.com/model-zoo/ImageNet_21K_P/models/timm/vit_base_patch16_224_in21k_miil.pth',
mean=(0, 0, 0), std=(1, 1, 1), crop_pct=0.875, interpolation='bilinear', num_classes=11221,
),
'vit_base_patch16_224_miil': _cfg(
url='https://miil-public-eu.oss-eu-central-1.aliyuncs.com/model-zoo/ImageNet_21K_P/models/timm'
'/vit_base_patch16_224_1k_miil_84_4.pth',
mean=(0, 0, 0), std=(1, 1, 1), crop_pct=0.875, interpolation='bilinear',
),
}
[docs]
class Attention(nn.Module):
def __init__(self, dim, num_heads=8, qkv_bias=False, attn_drop=0., proj_drop=0.):
super().__init__()
self.num_heads = num_heads
head_dim = dim // num_heads
self.scale = 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)
[docs]
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.unbind(0) # make torchscript happy (cannot use tensor as tuple)
attn = (q @ k.transpose(-2, -1)) * self.scale
attn = attn.softmax(dim=-1)
attn = self.attn_drop(attn)
x = (attn @ v).transpose(1, 2).reshape(B, N, C)
x = self.proj(x)
x = self.proj_drop(x)
return x
[docs]
class Block(nn.Module):
def __init__(self, dim, num_heads, mlp_ratio=4., qkv_bias=False, 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 = Attention(dim, num_heads=num_heads, qkv_bias=qkv_bias, 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)
[docs]
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
def _init_vit_weights(module: nn.Module, name: str = '', head_bias: float = 0., jax_impl: bool = False):
""" ViT weight initialization
* When called without n, head_bias, jax_impl args it will behave exactly the same
as my original init for compatibility with prev hparam / downstream use cases (ie DeiT).
* When called w/ valid n (module name) and jax_impl=True, will (hopefully) match JAX impl
"""
if isinstance(module, nn.Linear):
if name.startswith('head'):
nn.init.zeros_(module.weight)
nn.init.constant_(module.bias, head_bias)
elif name.startswith('pre_logits'):
lecun_normal_(module.weight)
nn.init.zeros_(module.bias)
else:
if jax_impl:
nn.init.xavier_uniform_(module.weight)
if module.bias is not None:
if 'mlp' in name:
nn.init.normal_(module.bias, std=1e-6)
else:
nn.init.zeros_(module.bias)
else:
trunc_normal_(module.weight, std=.02)
if module.bias is not None:
nn.init.zeros_(module.bias)
elif jax_impl and isinstance(module, nn.Conv2d):
# NOTE conv was left to pytorch default in my original init
lecun_normal_(module.weight)
if module.bias is not None:
nn.init.zeros_(module.bias)
elif isinstance(module, (nn.LayerNorm, nn.GroupNorm, nn.BatchNorm2d)):
nn.init.zeros_(module.bias)
nn.init.ones_(module.weight)
@torch.no_grad()
def _load_weights(model: VisionTransformer, checkpoint_path: str, prefix: str = ''):
""" Load weights from .npz checkpoints for official Google Brain Flax implementation
"""
import numpy as np
def _n2p(w, t=True):
if w.ndim == 4 and w.shape[0] == w.shape[1] == w.shape[2] == 1:
w = w.flatten()
if t:
if w.ndim == 4:
w = w.transpose([3, 2, 0, 1])
elif w.ndim == 3:
w = w.transpose([2, 0, 1])
elif w.ndim == 2:
w = w.transpose([1, 0])
return torch.from_numpy(w)
w = np.load(checkpoint_path)
if not prefix and 'opt/target/embedding/kernel' in w:
prefix = 'opt/target/'
if hasattr(model.patch_embed, 'backbone'):
# hybrid
backbone = model.patch_embed.backbone
stem_only = not hasattr(backbone, 'stem')
stem = backbone if stem_only else backbone.stem
stem.conv.weight.copy_(adapt_input_conv(stem.conv.weight.shape[1], _n2p(w[f'{prefix}conv_root/kernel'])))
stem.norm.weight.copy_(_n2p(w[f'{prefix}gn_root/scale']))
stem.norm.bias.copy_(_n2p(w[f'{prefix}gn_root/bias']))
if not stem_only:
for i, stage in enumerate(backbone.stages):
for j, block in enumerate(stage.blocks):
bp = f'{prefix}block{i + 1}/unit{j + 1}/'
for r in range(3):
getattr(block, f'conv{r + 1}').weight.copy_(_n2p(w[f'{bp}conv{r + 1}/kernel']))
getattr(block, f'norm{r + 1}').weight.copy_(_n2p(w[f'{bp}gn{r + 1}/scale']))
getattr(block, f'norm{r + 1}').bias.copy_(_n2p(w[f'{bp}gn{r + 1}/bias']))
if block.downsample is not None:
block.downsample.conv.weight.copy_(_n2p(w[f'{bp}conv_proj/kernel']))
block.downsample.norm.weight.copy_(_n2p(w[f'{bp}gn_proj/scale']))
block.downsample.norm.bias.copy_(_n2p(w[f'{bp}gn_proj/bias']))
embed_conv_w = _n2p(w[f'{prefix}embedding/kernel'])
else:
embed_conv_w = adapt_input_conv(
model.patch_embed.proj.weight.shape[1], _n2p(w[f'{prefix}embedding/kernel']))
model.patch_embed.proj.weight.copy_(embed_conv_w)
model.patch_embed.proj.bias.copy_(_n2p(w[f'{prefix}embedding/bias']))
model.cls_token.copy_(_n2p(w[f'{prefix}cls'], t=False))
pos_embed_w = _n2p(w[f'{prefix}Transformer/posembed_input/pos_embedding'], t=False)
if pos_embed_w.shape != model.pos_embed.shape:
pos_embed_w = resize_pos_embed( # resize pos embedding when different size from pretrained weights
pos_embed_w, model.pos_embed, getattr(model, 'num_tokens', 1), model.patch_embed.grid_size)
model.pos_embed.copy_(pos_embed_w)
model.norm.weight.copy_(_n2p(w[f'{prefix}Transformer/encoder_norm/scale']))
model.norm.bias.copy_(_n2p(w[f'{prefix}Transformer/encoder_norm/bias']))
if isinstance(model.head, nn.Linear) and model.head.bias.shape[0] == w[f'{prefix}head/bias'].shape[-1]:
model.head.weight.copy_(_n2p(w[f'{prefix}head/kernel']))
model.head.bias.copy_(_n2p(w[f'{prefix}head/bias']))
if isinstance(getattr(model.pre_logits, 'fc', None), nn.Linear) and f'{prefix}pre_logits/bias' in w:
model.pre_logits.fc.weight.copy_(_n2p(w[f'{prefix}pre_logits/kernel']))
model.pre_logits.fc.bias.copy_(_n2p(w[f'{prefix}pre_logits/bias']))
for i, block in enumerate(model.blocks.children()):
block_prefix = f'{prefix}Transformer/encoderblock_{i}/'
mha_prefix = block_prefix + 'MultiHeadDotProductAttention_1/'
block.norm1.weight.copy_(_n2p(w[f'{block_prefix}LayerNorm_0/scale']))
block.norm1.bias.copy_(_n2p(w[f'{block_prefix}LayerNorm_0/bias']))
block.attn.qkv.weight.copy_(torch.cat([
_n2p(w[f'{mha_prefix}{n}/kernel'], t=False).flatten(1).T for n in ('query', 'key', 'value')]))
block.attn.qkv.bias.copy_(torch.cat([
_n2p(w[f'{mha_prefix}{n}/bias'], t=False).reshape(-1) for n in ('query', 'key', 'value')]))
block.attn.proj.weight.copy_(_n2p(w[f'{mha_prefix}out/kernel']).flatten(1))
block.attn.proj.bias.copy_(_n2p(w[f'{mha_prefix}out/bias']))
for r in range(2):
getattr(block.mlp, f'fc{r + 1}').weight.copy_(_n2p(w[f'{block_prefix}MlpBlock_3/Dense_{r}/kernel']))
getattr(block.mlp, f'fc{r + 1}').bias.copy_(_n2p(w[f'{block_prefix}MlpBlock_3/Dense_{r}/bias']))
block.norm2.weight.copy_(_n2p(w[f'{block_prefix}LayerNorm_2/scale']))
block.norm2.bias.copy_(_n2p(w[f'{block_prefix}LayerNorm_2/bias']))
[docs]
def resize_pos_embed(posemb, posemb_new, num_tokens=1, gs_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 num_tokens:
posemb_tok, posemb_grid = posemb[:, :num_tokens], posemb[0, num_tokens:]
ntok_new -= num_tokens
else:
posemb_tok, posemb_grid = posemb[:, :0], posemb[0]
gs_old = int(math.sqrt(len(posemb_grid)))
if not len(gs_new): # backwards compatibility
gs_new = [int(math.sqrt(ntok_new))] * 2
assert len(gs_new) >= 2
_logger.info('Position embedding grid-size from %s to %s', [gs_old, 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, mode='bicubic', align_corners=False)
posemb_grid = posemb_grid.permute(0, 2, 3, 1).reshape(1, gs_new[0] * gs_new[1], -1)
posemb = torch.cat([posemb_tok, posemb_grid], dim=1)
return posemb
[docs]
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, getattr(model, 'num_tokens', 1), model.patch_embed.grid_size)
out_dict[k] = v
return out_dict
# def _create_vision_transformer(variant, pretrained=False, default_cfg=None, **kwargs):
# default_cfg = default_cfg or default_cfgs[variant]
# if kwargs.get('features_only', None):
# raise RuntimeError('features_only not implemented for Vision Transformer models.')
# # NOTE this extra code to support handling of repr size for in21k pretrained models
# default_num_classes = default_cfg['num_classes']
# num_classes = kwargs.get('num_classes', default_num_classes)
# 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 = build_model_with_cfg(
# VisionTransformer, variant, pretrained,
# default_cfg=default_cfg,
# representation_size=repr_size,
# pretrained_filter_fn=checkpoint_filter_fn,
# pretrained_custom_load='npz' in default_cfg['url'],
# **kwargs)
# return model
def _create_vision_transformer(variant, pretrained=False, **kwargs):
if kwargs.get('features_only', None):
raise RuntimeError('features_only not implemented for Vision Transformer models.')
if 'flexi' in variant:
# FIXME Google FlexiViT pretrained models have a strong preference for bilinear patch / embed
# interpolation, other pretrained models resize better w/ anti-aliased bicubic interpolation.
_filter_fn = partial(checkpoint_filter_fn, interpolation='bilinear', antialias=False)
else:
_filter_fn = checkpoint_filter_fn
# FIXME attn pool (currently only in siglip) params removed if pool disabled, is there a better soln?
strict = True
if 'siglip' in variant and kwargs.get('global_pool', None) != 'map':
strict = False
pretrained_cfg = resolve_pretrained_cfg(variant, pretrained_cfg=kwargs.pop('pretrained_cfg', None))
pretrained_cfg.custom_load = True
return build_model_with_cfg(
VisionTransformer,
variant,
pretrained,
pretrained_cfg=pretrained_cfg,
pretrained_filter_fn=_filter_fn,
pretrained_strict=strict,
**kwargs,
)
[docs]
def vit_base_patch16_224_in21k(pretrained=False, adapter=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.
NOTE: this model has valid 21k classifier head and no representation (pre-logits) layer
"""
model_kwargs = dict(
patch_size=16, embed_dim=768, depth=12, num_heads=12, with_adapter=adapter, **kwargs)
model = _create_vision_transformer('vit_base_patch16_224_in21k', pretrained=pretrained, **model_kwargs)
del model.head
del model.norm
model.norm = nn.LayerNorm(768)
return model
[docs]
def vit_base_patch16_224_mocov3(pretrained=False, adapter=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.
NOTE: this model has valid 21k classifier head and no representation (pre-logits) layer
"""
model_kwargs = dict(
patch_size=16, embed_dim=768, depth=12, num_heads=12, with_adapter=adapter, **kwargs)
model = _create_vision_transformer('vit_base_patch16_224_in21k', pretrained=False, **model_kwargs)
del model.head
ckpt = torch.load('mocov3-vit-base-300ep.pth', map_location='cpu')['model']
state_dict = model.state_dict()
state_dict.update(ckpt)
model.load_state_dict(state_dict)
del model.norm
model.norm = nn.LayerNorm(768)
return model