官方提供的代碼：https://github.com/google-research/vision_transformer
大佬複現的版本：https://github.com/lucidrains/vit-pytorch
對不起，我好菜，官方給的代碼我确實看不懂啊，是以看了第二個版本的代碼。第二個版本的代碼超級受歡迎且易使用，我看的時候，Git repo已經被star 5.7k次。大家直接
pip install vit-pytorch
就好。
是以作為初次接觸vit的同學們來說，推薦看第二個版本，結構清晰明了。
1. 大佬複現版本給的使用案例
import torch
from vit_pytorch import ViT
v = ViT(
image_size = 256, # 圖像大小
patch_size = 32, # patch大小（分塊的大小）
num_classes = 1000, # imagenet資料集1000分類
dim = 1024, # position embedding的次元
depth = 6, # encoder和decoder中block層數是6
heads = 16, # multi-head中head的數量為16
mlp_dim = 2048,
dropout = 0.1, #
emb_dropout = 0.1
)
img = torch.randn(1, 3, 256, 256)
preds = v(img) # (1, 1000)
大家完全可以把這段代碼copy-paste到自己的pycharm裡，然後使用調試功能，一步步看ViT的每一步操作。
2. Transformer結構
進行6次for循環，有6層encoder和decoder結構。for循環内部依次使用multi-head attention和Feed Forward，對應Transformer的Encoder結構中多頭自注意力子產品和MLP子產品。在自注意力後及feed forward後，有殘差連接配接。
class Transformer(nn.Module):
def __init__(self, dim, depth, heads, dim_head, mlp_dim, dropout = 0.):
super().__init__()
self.layers = nn.ModuleList([])
for _ in range(depth):
self.layers.append(nn.ModuleList([
PreNorm(dim, Attention(dim, heads = heads, dim_head = dim_head, dropout = dropout)),
PreNorm(dim, FeedForward(dim, mlp_dim, dropout = dropout))
]))
def forward(self, x):
for attn, ff in self.layers:
x = attn(x) + x
x = ff(x) + x
return x
PreNorm類代碼如下，在使用multi-head attention和Feed Forward之前，首先對輸入通過LayerNorm進行處理。
class PreNorm(nn.Module):
def __init__(self, dim, fn):
super().__init__()
self.norm = nn.LayerNorm(dim)
self.fn = fn
def forward(self, x, **kwargs):
return self.fn(self.norm(x), **kwargs)
可以參考論文中的圖：
3. Attention
class Attention(nn.Module):
def __init__(self, dim, heads = 8, dim_head = 64, dropout = 0.):
super().__init__()
inner_dim = dim_head * heads
project_out = not (heads == 1 and dim_head == dim)
self.heads = heads
self.scale = dim_head ** -0.5
self.attend = nn.Softmax(dim = -1)
self.to_qkv = nn.Linear(dim, inner_dim * 3, bias = False)
self.to_out = nn.Sequential(
nn.Linear(inner_dim, dim),
nn.Dropout(dropout)
) if project_out else nn.Identity()
def forward(self, x):
qkv = self.to_qkv(x).chunk(3, dim = -1) # 首先生成q,k,v
q, k, v = map(lambda t: rearrange(t, 'b n (h d) -> b h n d', h = self.heads), qkv)
dots = torch.matmul(q, k.transpose(-1, -2)) * self.scale
attn = self.attend(dots)
out = torch.matmul(attn, v)
out = rearrange(out, 'b h n d -> b n (h d)')
return self.to_out(out)
torch.chunk(tensor, chunk_num, dim)函數的功能：與torch.cat()剛好相反，它是将tensor按dim（行或列）分割成chunk_num個tensor塊，傳回的是一個元組。
attention操作的整體流程：
- 首先對輸入生成query, key和value，這裡的“輸入”有可能是整個網絡的輸入，也可能是某個hidden layer的output。在這裡，生成的qkv是個長度為3的元組，每個元組的大小為(1, 65, 1024)
- 對qkv進行處理，重新指定次元，得到的q, k, v次元均為(1, 16, 65, 64)
- q和k做點乘，得到的dots次元為(1, 16, 65, 65)
- 對dots的最後一維做softmax，得到各個patch對其他patch的注意力得分
- 将attention和value做點乘
- 對各個次元重新排列，得到與輸入相同次元的輸出 (1, 65, 1024)
4. FeedForward
class FeedForward(nn.Module):
def __init__(self, dim, hidden_dim, dropout = 0.):
super().__init__()
self.net = nn.Sequential(
nn.Linear(dim, hidden_dim), # dim=1024, hidden_dim=2048
nn.GELU(),
nn.Dropout(dropout),
nn.Linear(hidden_dim, dim),
nn.Dropout(dropout)
)
def forward(self, x):
return self.net(x)
FeedForward子產品共有2個全連接配接層，整個結構是：
- 首先過一個全連接配接層
- 經過GELU()激活函數進行處理
- nn.Dropout()，以一定機率丢失掉一些神經元，防止過拟合
- 再過一個全連接配接層
-
nn.Dropout()
注意：GELU(x) = x * Φ(x), 其中，Φ(x)是高斯分布的累積分布函數 。
5. ViT操作流程
ViT的各個結構都寫在了__init__()裡，不再細講，通過forward()來看ViT的整個前向傳播過程（操作流程）。
class ViT(nn.Module):
def __init__(self, *, image_size, patch_size, num_classes, dim, depth, heads, mlp_dim, pool = 'cls', channels = 3, dim_head = 64, dropout = 0., emb_dropout = 0.):
super().__init__()
image_height, image_width = pair(image_size)
patch_height, patch_width = pair(patch_size)
assert image_height % patch_height == 0 and image_width % patch_width == 0, 'Image dimensions must be divisible by the patch size.'
num_patches = (image_height // patch_height) * (image_width // patch_width)
patch_dim = channels * patch_height * patch_width
assert pool in {'cls', 'mean'}, 'pool type must be either cls (cls token) or mean (mean pooling)'
self.to_patch_embedding = nn.Sequential(
Rearrange('b c (h p1) (w p2) -> b (h w) (p1 p2 c)', p1 = patch_height, p2 = patch_width),
nn.Linear(patch_dim, dim),
)
self.pos_embedding = nn.Parameter(torch.randn(1, num_patches + 1, dim)) # (1,65,1024)
self.cls_token = nn.Parameter(torch.randn(1, 1, dim))
self.dropout = nn.Dropout(emb_dropout)
self.transformer = Transformer(dim, depth, heads, dim_head, mlp_dim, dropout)
self.pool = pool
self.to_latent = nn.Identity()
self.mlp_head = nn.Sequential(
nn.LayerNorm(dim),
nn.Linear(dim, num_classes)
)
def forward(self, img): # img: (1, 3, 256, 256)
x = self.to_patch_embedding(img) # (1, 64, 1024)
b, n, _ = x.shape
cls_tokens = repeat(self.cls_token, '() n d -> b n d', b = b) # (1, 1, 1024)
x = torch.cat((cls_tokens, x), dim=1) # (1, 65, 1024)
x += self.pos_embedding[:, :(n + 1)] # (1, 65, 1024)
x = self.dropout(x) # (1, 65, 1024)
x = self.transformer(x) # (1, 65, 1024)
x = x.mean(dim = 1) if self.pool == 'mean' else x[:, 0] # (1, 1024)
x = self.to_latent(x)
return self.mlp_head(x)
整體流程：
- 首先對輸入進來的img(256*256大小)，劃分為32*32大小的patch，共有8*8個。并将patch轉換成embedding。（對應第26行代碼）
- 生成cls_tokens （對應第28行代碼）
- 将cls_tokens沿dim=1維與x進行拼接 （對應第29行代碼）
- 生成随機的position embedding，每個embedding都是1024維 （對應代碼14行和30行）
- 對輸入經過Transformer進行編碼（對應代碼第32行）
- 如果是分類任務的話，截取第一個可學習的class embedding
- 最後過一個MLP Head用于分類。
![考證大全 | 證券從業資格考試[圖]](data:image/gif;base64,R0lGODlhAQABAIAAAP///wAAACwAAAAAAQABAAACAkQBADs=)