43 lines
1.1 KiB
Python
43 lines
1.1 KiB
Python
# Copyright (c) Facebook, Inc. and its affiliates.
|
|
# All rights reserved.
|
|
|
|
# This source code is licensed under the license found in the
|
|
# LICENSE file in the root directory of this source tree.
|
|
|
|
import math
|
|
import torch
|
|
import torch.nn as nn
|
|
from functools import partial, reduce
|
|
from operator import mul
|
|
|
|
from timm.models.vision_transformer import VisionTransformer, _cfg
|
|
|
|
__all__ = [
|
|
'vit_small',
|
|
'vit_base',
|
|
]
|
|
|
|
|
|
def vit_small(**kwargs):
|
|
model = VisionTransformer(
|
|
patch_size=16, embed_dim=384, depth=12, num_heads=12, mlp_ratio=4, qkv_bias=True, num_classes=256,
|
|
norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
|
|
# model.default_cfg = _cfg()
|
|
return model
|
|
|
|
|
|
def vit_base(**kwargs):
|
|
model = VisionTransformer(
|
|
patch_size=16, embed_dim=768, depth=12, num_heads=12, mlp_ratio=4, qkv_bias=True, num_classes=256,
|
|
norm_layer=partial(nn.LayerNorm, eps=1e-6), **kwargs)
|
|
model.default_cfg = _cfg(num_classes=256)
|
|
return model
|
|
|
|
|
|
if __name__ == '__main__':
|
|
img = torch.randn(8, 3, 224, 224)
|
|
vit = vit_base()
|
|
out = vit(img)
|
|
print(out.shape)
|
|
# print(count_parameters(vit))
|