代码收藏家 Yolov5目标检测中的Transformer模型详解（Python实现）
我们知道，transfomer模型常用于自然语言处理、大模型、图像生成等场景，我在另一篇博客2.1 transformer模型原理及代码(python)-CSDN博客通过一个很简单的例子博客详细介绍了transfomer模型的每个模块的功能及应用，方便我们认识transfomer模型，这篇博文通过yolov5的trabnsformer模型对该模型的每个模块做一个介绍，这样我们能从图形识别的角度更深的了解transformer模型。
1. transformer模型及运行结果
代码如下：
import math
import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.utils.checkpoint as checkpoint
# 1.截断正态分布，用来初始化权重张量（用截断正态分布填充张量，a是下限，b是上限，mean是均值,正态分布的标准差,tensor要初始化的张量）
def trunc_normal_(tensor, mean=0., std=1., a=-2., b=2.):
def _no_grad_trunc_normal_(tensor, mean, std, a, b):
def norm_cdf(x):
return (1. + math.erf(x / math.sqrt(2.))) / 2.
with torch.no_grad():
l = norm_cdf((a - mean) / std)
u = norm_cdf((b - mean) / std)
tensor.uniform_(2 * l - 1, 2 * u - 1)
tensor.erfinv_()
tensor.mul_(std * math.sqrt(2.))
tensor.add_(mean)
tensor.clamp_(min=a, max=b)
return tensor
return _no_grad_trunc_normal_(tensor, mean, std, a, b)
# 2.对输入进来的图片进行高和宽的压缩
class PatchEmbed(nn.Module):
def __init__(self, img_size=[224, 224], patch_size=4, in_chans=3, embed_dim=96, norm_layer=None):
super().__init__()
# [224, 224]
self.img_size = img_size
# [4, 4]
self.patch_size = [patch_size, patch_size]
# [56, 56]
self.patches_resolution = [self.img_size[0] // self.patch_size[0], self.img_size[1] // self.patch_size[1]]
# 3
self.in_chans = in_chans
# 96
self.embed_dim = embed_dim
# 3136
self.num_patches = self.patches_resolution[0] * self.patches_resolution[1]
# -------------------------------------------------------#
# bs, 224, 224, 3 -> bs, 56, 56, 96
# -------------------------------------------------------#
self.proj = nn.Conv2d(in_chans,embed_dim,kernel_size=patch_size, stride=patch_size)# 卷积结构
if norm_layer is not None:
self.norm = norm_layer(embed_dim)
else:
self.norm = None
def forward(self,x):
# x = self.proj(x) # bs, 3, 224, 224 -> bs, 96, 56, 56
# x = x.flatten(2) # bs, 96, 56, 56 -> 4, 96, 3136
# x = x.transpose(1, 2) # 4, 96, 3136 -> 4, 3136, 96
x = self.proj(x).flatten(2).transpose(1, 2)
if self.norm is not None:
x = self.norm(x)
return x
# 3. 确保网络层的通道数可以被divisor（通常是8）整除： v:原始通道数； divisor：除数，通常是8； min_value：通道数的最小值。
def _make_divisible(v, divisor, min_value=None):
if min_value is None:
min_value = divisor
new_v = max(min_value, int(v + divisor / 2) // divisor * divisor)
if new_v < 0.9 * v:
new_v += divisor
return new_v
# 4. 将图像分割成小块以便进行特征提取和处理
def window_partition(x, window_size):
B, H, W, C = x.shape
# ------------------------------------------------------------------#
# bs, 56, 56, 96 -> bs, 8, 7, 8, 7, 96 -> bs * 64, 7, 7, 96
# ------------------------------------------------------------------#
x = x.view(B, H // window_size, window_size, W // window_size, window_size, C)
windows = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(-1, window_size, window_size, C)
return windows
# 5.激活函数GELU
class GELU(nn.Module):
def __init__(self):
super(GELU, self).__init__()
def forward(self,x):
return 0.5 * x * (1 + torch.tanh(np.sqrt(2 / np.pi) * (x + 0.044715 * torch.pow(x, 3))))
# 6.对输入进来的特征层进行高和宽的压缩
# -------------------------------------------------------#
# 对输入进来的特征层进行高和宽的压缩
# 进行跨特征点的特征提取，提取完成后进行堆叠。
# -------------------------------------------------------#
class PatchMerging(nn.Module):
def __init__(self, input_resolution, dim, norm_layer=nn.LayerNorm):
super().__init__()
self.input_resolution = input_resolution
self.dim = dim
self.norm = norm_layer(4 * dim)
self.reduction = nn.Linear(4 * dim, 2 * dim, bias=False)
def forward(self, x):
H, W = self.input_resolution
B, L, C = x.shape
assert L == H * W, "input feature has wrong size"
assert H % 2 == 0 and W % 2 == 0, f"x size ({H}*{W}) are not even."
# -------------------------------------------------------#
# bs, 3136, 96 -> bs, 56, 56, 96
# -------------------------------------------------------#
x = x.view(B, H, W, C)
# -------------------------------------------------------#
# x0 ~ x3 bs, 56, 56, 96 -> bs, 28, 28, 96
# -------------------------------------------------------#
x0 = x[:, 0::2, 0::2, :] # B H/2 W/2 C
x1 = x[:, 1::2, 0::2, :] # B H/2 W/2 C
x2 = x[:, 0::2, 1::2, :] # B H/2 W/2 C
x3 = x[:, 1::2, 1::2, :] # B H/2 W/2 C
# -------------------------------------------------------#
# 4 X bs, 28, 28, 96 -> bs, 28, 28, 384
# -------------------------------------------------------#
x = torch.cat([x0, x1, x2, x3], -1) # B H/2 W/2 4*C
# -------------------------------------------------------#
# bs, 28, 28, 384 -> bs, 784, 384
# -------------------------------------------------------#
x = x.view(B, -1, 4 * C) # B H/2*W/2 4*C
# -------------------------------------------------------#
# bs, 784, 384 -> bs, 784, 192
# -------------------------------------------------------#
x = self.norm(x)
x = self.reduction(x)
return x
# 7.注意力机制：将序列划分为多个窗口，每个窗口内的元素进行注意力计算，从而减少计算量和显存需求
class WindowAttention(nn.Module):
def __init__(self, dim, window_size, num_heads, qkv_bias=True, qk_scale=None, attn_drop=0., proj_drop=0.):
super().__init__()
self.dim = dim
self.window_size = window_size # Wh, Ww
self.num_heads = num_heads
head_dim = dim // num_heads
self.scale = qk_scale or head_dim ** -0.5
# --------------------------------------------------------------------------#
# 相对坐标矩阵，用于表示每个窗口内，其它点相对于自己的坐标
# 由于相对坐标取值范围为-6 ~ +6。中间共13个值，因此需要13 * 13
# 13 * 13, num_heads
# --------------------------------------------------------------------------#
self.relative_position_bias_table = nn.Parameter(
torch.zeros((2 * window_size[0] - 1) * (2 * window_size[1] - 1), num_heads)
)
# --------------------------------------------------------------------------#
# 该部分用于获取7x7的矩阵内部，其它特征点相对于自身相对坐标
# --------------------------------------------------------------------------#
coords_h = torch.arange(self.window_size[0])
coords_w = torch.arange(self.window_size[1])
coords = torch.stack(torch.meshgrid([coords_h, coords_w])) # 2, Wh, Ww
coords_flatten = torch.flatten(coords, 1) # 2, Wh*Ww
relative_coords = coords_flatten[:, :, None] - coords_flatten[:, None, :] # 2, Wh*Ww, Wh*Ww
relative_coords = relative_coords.permute(1, 2, 0).contiguous() # Wh*Ww, Wh*Ww, 2
relative_coords[:, :, 0] += self.window_size[0] - 1 # shift to start from 0
relative_coords[:, :, 1] += self.window_size[1] - 1
relative_coords[:, :, 0] *= 2 * self.window_size[1] - 1
relative_position_index = relative_coords.sum(-1) # Wh*Ww, Wh*Ww
self.register_buffer("relative_position_index", relative_position_index)
# --------------------------------------------------------------------------#
# 乘积获得q、k、v，用于计算多头注意力机制
# --------------------------------------------------------------------------#
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)
trunc_normal_(self.relative_position_bias_table, std=.02)
self.softmax = nn.Softmax(dim=-1)
def forward(self, x, mask=None):
B_, N, C = x.shape
# --------------------------------------------------------------------------#
# bs * 64, 49, 96 -> bs * 64, 49, 96 * 3 ->
# bs * 64, 49, 3, num_heads, 32 -> 3, bs * 64, num_head, 49, 32
# --------------------------------------------------------------------------#
qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4)
# --------------------------------------------------------------------------#
# bs * 64, num_head, 49, 32
# --------------------------------------------------------------------------#
q, k, v = qkv[0], qkv[1], qkv[2]
# --------------------------------------------------------------------------#
# bs * 64, num_head, 49, 49
# --------------------------------------------------------------------------#
q = q * self.scale
attn = (q @ k.transpose(-2, -1))
# --------------------------------------------------------------------------#
# 这一步是根据已经求得的注意力，加上相对坐标的偏执量
# 形成最后的注意力
# --------------------------------------------------------------------------#
relative_position_bias = self.relative_position_bias_table[self.relative_position_index.view(-1)].view(
self.window_size[0] * self.window_size[1], self.window_size[0] * self.window_size[1], -1) # Wh*Ww,Wh*Ww,nH
relative_position_bias = relative_position_bias.permute(2, 0, 1).contiguous()
attn = attn + relative_position_bias.unsqueeze(0)
# --------------------------------------------------------------------------#
# 加上mask，保证分区。
# bs * 64, num_head, 49, 49
# --------------------------------------------------------------------------#
if mask is not None:
nW = mask.shape[0]
attn = attn.view(B_ // nW, nW, self.num_heads, N, N) + mask.unsqueeze(1).unsqueeze(0)
attn = attn.view(-1, self.num_heads, N, N)
attn = self.softmax(attn)
else:
attn = self.softmax(attn)
attn = self.attn_drop(attn)
# ---------------------------------------------------------------------------------------#
# bs * 64, num_head, 49, 49 @ bs * 64, num_head, 49, 32 -> bs * 64, num_head, 49, 32
#
# bs * 64, num_head, 49, 32 -> bs * 64, 49, 96
# ---------------------------------------------------------------------------------------#
x = (attn @ v).transpose(1, 2).reshape(B_, N, C)
x = self.proj(x)
x = self.proj_drop(x)
return x
# 8.生成一系列的mask来选择网络中的分支。mask为1的地方，保留相应的网络结构；mask为0的地方，使该部分网络结构失效
def drop_path(x, drop_prob: float = 0., training: bool = False, scale_by_keep: bool = True):
"""
Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).
This is the same as the DropConnect impl I created for EfficientNet, etc networks, however,
the original name is misleading as 'Drop Connect' is a different form of dropout in a separate paper...
See discussion: https://github.com/tensorflow/tpu/issues/494#issuecomment-532968956 ... I've opted for
changing the layer and argument names to 'drop path' rather than mix DropConnect as a layer name and use
'survival rate' as the argument.
"""
if drop_prob == 0. or not training:
return x
keep_prob = 1 - drop_prob
shape = (x.shape[0],) + (1,) * (x.ndim - 1) # work with diff dim tensors, not just 2D ConvNets
random_tensor = x.new_empty(shape).bernoulli_(keep_prob)
if keep_prob > 0.0 and scale_by_keep:
random_tensor.div_(keep_prob)
return x * random_tensor
class DropPath(nn.Module):
"""
Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).
"""
def __init__(self, drop_prob=None, scale_by_keep=True):
super(DropPath, self).__init__()
self.drop_prob = drop_prob
self.scale_by_keep = scale_by_keep
def forward(self, x):
return drop_path(x, self.drop_prob, self.training, self.scale_by_keep)
# 9. 两次全连接
class Mlp(nn.Module):
def __init__(self, in_features, hidden_features=None, out_features=None, act_layer=GELU, drop=0.):
super().__init__()
out_features = out_features or in_features
hidden_features = hidden_features or in_features
self.fc1 = nn.Linear(in_features, hidden_features)
self.act = act_layer()
self.fc2 = nn.Linear(hidden_features, out_features)
self.drop = nn.Dropout(drop)
def forward(self, x):
x = self.fc1(x)
x = self.act(x)
x = self.drop(x)
x = self.fc2(x)
x = self.drop(x)
return x
# 10.将窗口内的信息重新组合回原始的特征图或图像
def window_reverse(windows, window_size, H, W):
# ------------------------------------------------------------------#
# bs * 64, 7, 7, 96 -> bs, 8, 8, 7, 7, 96 -> bs, 56, 56, 96
# ------------------------------------------------------------------#
B = int(windows.shape[0] / (H * W / window_size / window_size))
x = windows.view(B, H // window_size, W // window_size, window_size, window_size, -1)
x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1)
return x
# 11.每个阶段重复的基础模块
# -------------------------------------------------------#
# 每个阶段重复的基础模块
# 在这其中会使用WindowAttention进行特征提取
# -------------------------------------------------------#
class SwinTransformerBlock(nn.Module):
def __init__(self, dim, input_resolution, num_heads, window_size=7, shift_size=0,
mlp_ratio=4., qkv_bias=True, qk_scale=None, drop=0., attn_drop=0., drop_path=0.,
act_layer=GELU, norm_layer=nn.LayerNorm):
super().__init__()
self.dim = dim
self.input_resolution = input_resolution
self.num_heads = num_heads
self.window_size = window_size
self.shift_size = shift_size
self.mlp_ratio = mlp_ratio
if min(self.input_resolution) <= self.window_size:
self.shift_size = 0
self.window_size = min(self.input_resolution)
assert 0 <= self.shift_size < self.window_size, "shift_size must in 0-window_size"
self.norm1 = norm_layer(dim)
self.attn = WindowAttention(
dim,
window_size=[self.window_size, self.window_size],
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)
mlp_hidden_dim = int(dim * mlp_ratio)
self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop)
if self.shift_size > 0:
# ----------------------------------------------------------------#
# 由于进行特征提取时，会对输入的特征层进行的平移
# 如：
# [ [
# [1, 2, 3], [5, 6, 4],
# [4, 5, 6], --> [8, 9, 7],
# [7, 8, 9], [1, 2, 3],
# ] ]
# 这一步的作用就是使得平移后的区域块只计算自己部分的注意力机制
# ----------------------------------------------------------------#
H, W = self.input_resolution
_H, _W = _make_divisible(H, self.window_size), _make_divisible(W, self.window_size),
img_mask = torch.zeros((1, _H, _W, 1)) # 1 H W 1
h_slices = (slice(0, -self.window_size),
slice(-self.window_size, -self.shift_size),
slice(-self.shift_size, None))
w_slices = (slice(0, -self.window_size),
slice(-self.window_size, -self.shift_size),
slice(-self.shift_size, None))
cnt = 0
for h in h_slices:
for w in w_slices:
img_mask[:, h, w, :] = cnt
cnt += 1
mask_windows = window_partition(img_mask, self.window_size) # nW, window_size, window_size, 1
mask_windows = mask_windows.view(-1, self.window_size * self.window_size)
attn_mask = mask_windows.unsqueeze(1) - mask_windows.unsqueeze(2)
attn_mask = attn_mask.masked_fill(attn_mask != 0, float(-100.0)).masked_fill(attn_mask == 0, float(0.0))
self.attn_mask = attn_mask.cpu().numpy()
else:
self.attn_mask = None
def forward(self, x):
H, W = self.input_resolution
B, L, C = x.shape
# assert L == H * W, "input feature has wrong size"
# -----------------------------------------------#
# bs, 3136, 96 -> bs, 56, 56, 96
# -----------------------------------------------#
shortcut = x
x = self.norm1(x)
x = x.view(B, H, W, C)
_H, _W = _make_divisible(H, self.window_size), _make_divisible(W, self.window_size),
x = x.permute(0, 3, 1, 2)
x = F.interpolate(x, [_H, _W], mode='bicubic', align_corners=False).permute(0, 2, 3, 1)
# -----------------------------------------------#
# 进行特征层的平移
# -----------------------------------------------#
if self.shift_size > 0:
shifted_x = torch.roll(x, shifts=(-self.shift_size, -self.shift_size), dims=(1, 2))
else:
shifted_x = x
# ------------------------------------------------------------------------------------------#
# bs, 56, 56, 96 -> bs * 64, 7, 7, 96 -> bs * 64, 49, 96
# ------------------------------------------------------------------------------------------#
x_windows = window_partition(shifted_x, self.window_size) # num_windows * B, window_size, window_size, C
x_windows = x_windows.view(-1, self.window_size * self.window_size, C) # nW*B, window_size*window_size, C
# -----------------------------------------------#
# bs * 64, 49, 97 -> bs * 64, 49, 97
# -----------------------------------------------#
if type(self.attn_mask) != type(None):
attn_mask = torch.tensor(self.attn_mask).cuda() if x.is_cuda else torch.tensor(self.attn_mask)
else:
attn_mask = None
attn_windows = self.attn(x_windows, mask=attn_mask) # nW*B, window_size*window_size, C
# -----------------------------------------------#
# bs * 64, 49, 97 -> bs, 56, 56, 96
# -----------------------------------------------#
attn_windows = attn_windows.view(-1, self.window_size, self.window_size, C)
shifted_x = window_reverse(attn_windows, self.window_size, _H, _W) # B H' W' C
# -----------------------------------------------#
# 将特征层平移回来
# -----------------------------------------------#
if self.shift_size > 0:
x = torch.roll(shifted_x, shifts=(self.shift_size, self.shift_size), dims=(1, 2))
else:
x = shifted_x
x = x.permute(0, 3, 1, 2)
x = F.interpolate(x, [H, W], mode='bicubic', align_corners=False).permute(0, 2, 3, 1)
# -----------------------------------------------#
# bs, 3136, 96
# -----------------------------------------------#
x = x.view(B, H * W, C)
# -----------------------------------------------#
# FFN
# bs, 3136, 96
# -----------------------------------------------#
x = shortcut + self.drop_path(x)
x = x + self.drop_path(self.mlp(self.norm2(x)))
return x
# 12.Swin-Transformer的基础模块
# -------------------------------------------------------#
# Swin-Transformer的基础模块。
# 使用窗口多头注意力机制进行特征提取。
# 使用PatchMerging进行高和宽的压缩。
# -------------------------------------------------------#
class BasicLayer(nn.Module):
def __init__(self, dim, input_resolution, depth, num_heads, window_size,
mlp_ratio=4., qkv_bias=True, qk_scale=None, drop=0., attn_drop=0.,
drop_path=0., norm_layer=nn.LayerNorm, downsample=None, use_checkpoint=False):
super().__init__()
# -------------------------------------------------------#
# 四个阶段对应不同的dim
# [96, 192, 384, 768]
# -------------------------------------------------------#
self.dim = dim
# -------------------------------------------------------#
# 四个阶段对应不同的输入分辨率
# [[56, 56], [28, 28], [14, 14], [7, 7]]
# -------------------------------------------------------#
self.input_resolution = input_resolution
# -------------------------------------------------------#
# 四个阶段对应不同的多头注意力机制重复次数
# [2, 2, 6, 2]
# -------------------------------------------------------#
self.depth = depth
self.use_checkpoint = use_checkpoint
# -------------------------------------------------------#
# 根据depth的次数利用窗口多头注意力机制进行特征提取。
# -------------------------------------------------------#
self.blocks = nn.ModuleList(
[
SwinTransformerBlock(
dim=dim,
input_resolution=input_resolution,
num_heads=num_heads,
window_size=window_size,
shift_size=0 if (i % 2 == 0) else window_size // 2,
mlp_ratio=mlp_ratio,
qkv_bias=qkv_bias,
qk_scale=qk_scale,
drop=drop,
attn_drop=attn_drop,
drop_path=drop_path[i] if isinstance(drop_path, list) else drop_path,
norm_layer=norm_layer
)
for i in range(depth)
]
)
if downsample is not None:
# -------------------------------------------------------#
# 判断是否要进行下采样，即：高宽压缩
# -------------------------------------------------------#
self.downsample = downsample(input_resolution, dim=dim, norm_layer=norm_layer)
else:
self.downsample = None
def forward(self, x):
for blk in self.blocks:
if self.use_checkpoint:
x_ = checkpoint.checkpoint(blk, x)
else:
x_ = blk(x)
if self.downsample is not None:
x = self.downsample(x_)
else:
x = x_
return x_, x
class SwinTransformer(nn.Module):
def __init__(self, img_size=[640, 640], patch_size=4, in_chans=3, num_classes=1000,
embed_dim=96, depths=[2, 2, 6, 2], num_heads=[3, 6, 12, 24],
window_size=7, mlp_ratio=4., qkv_bias=True, qk_scale=None,
drop_rate=0., attn_drop_rate=0., drop_path_rate=0.1,
norm_layer=nn.LayerNorm, ape=False, patch_norm=True,
use_checkpoint=False, **kwargs):
super().__init__()
self.num_classes = num_classes
self.num_layers = len(depths)
self.embed_dim = embed_dim
self.ape = ape
self.patch_norm = patch_norm
self.num_features = int(embed_dim * 2 ** (self.num_layers - 1))
self.mlp_ratio = mlp_ratio
# --------------------------------------------------#
# bs, 224, 224, 3 -> bs, 3136, 96
# --------------------------------------------------#
self.patch_embed = PatchEmbed(
img_size=img_size,
patch_size=patch_size,
in_chans=in_chans,
embed_dim=embed_dim,
norm_layer=norm_layer if self.patch_norm else None
)
# --------------------------------------------------#
# PatchEmbed之后的图像序列长度 3136
# PatchEmbed之后的图像对应的分辨率 [56, 56]
# --------------------------------------------------#
num_patches = self.patch_embed.num_patches
patches_resolution = self.patch_embed.patches_resolution
self.patches_resolution = patches_resolution
if self.ape:
self.absolute_pos_embed = nn.Parameter(torch.zeros(1, num_patches, embed_dim))
trunc_normal_(self.absolute_pos_embed, std=.02)
self.pos_drop = nn.Dropout(p=drop_rate)
# --------------------------------------------------#
# stochastic depth
# --------------------------------------------------#
dpr = [x.item() for x in torch.linspace(0, drop_path_rate, sum(depths))] # stochastic depth decay rule
# ---------------------------------------------------------------#
# 构建swin-transform的每个阶段
# bs, 3136, 96 -> bs, 784, 192 -> bs, 196, 384 -> bs, 49, 768
# ---------------------------------------------------------------#
self.layers = nn.ModuleList()
for i_layer in range(self.num_layers):
layer = BasicLayer(
dim=int(embed_dim * 2 ** i_layer),
input_resolution=(patches_resolution[0] // (2 ** i_layer), patches_resolution[1] // (2 ** i_layer)),
depth=depths[i_layer],
num_heads=num_heads[i_layer],
window_size=window_size,
mlp_ratio=self.mlp_ratio,
qkv_bias=qkv_bias,
qk_scale=qk_scale,
drop=drop_rate,
attn_drop=attn_drop_rate,
drop_path=dpr[sum(depths[:i_layer]):sum(depths[:i_layer + 1])],
norm_layer=norm_layer,
downsample=PatchMerging if (i_layer < self.num_layers - 1) else None,
use_checkpoint=use_checkpoint
)
self.layers.append(layer)
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 {'absolute_pos_embed'}
@torch.jit.ignore
def no_weight_decay_keywords(self):
return {'relative_position_bias_table'}
def forward(self, x):
x = self.patch_embed(x)
if self.ape:
x = x + self.absolute_pos_embed
x = self.pos_drop(x)
inverval_outs = []
for i, layer in enumerate(self.layers):
x_, x = layer(x)
if i != 0:
inverval_outs.append(x_)
outs = []
for i, layer in enumerate(inverval_outs):
H, W = (self.patches_resolution[0] // (2 ** (i + 1)), self.patches_resolution[1] // (2 ** (i + 1)))
B, L, C = layer.shape
layer = layer.view([B, H, W, C]).permute([0, 3, 1, 2])
outs.append(layer)
return outs
def Swin_transformer_Tiny(pretrained=False, input_shape=[640, 640], **kwargs):
model = SwinTransformer(input_shape, depths=[2, 2, 6, 2], **kwargs)
if pretrained:
url = "https://github.com/bubbliiiing/yolov5-pytorch/releases/download/v1.0/swin_tiny_patch4_window7.pth"
checkpoint = torch.hub.load_state_dict_from_url(url=url, map_location="cpu", model_dir="./model_data")
model.load_state_dict(checkpoint, strict=False)
print("Load weights from ", url.split('/')[-1])
return model
if __name__ == "__main__":
datas = torch.randn(4,3,640,640)
nets = SwinTransformer()
outs = nets(datas)
print("\n",".......... outs如下：............")
for item in outs:
print(item.size())
运行结果如下：
在该模型中输入的参数如下：
img_size : [640,640]
datas : (4,3,640,640)，表示图片的大小为[640, 640, 3], Batch为4.
patch_size: 4，将图片的尺寸缩小到1/4
in_chans：3，初始输入图片的通道数；一般彩图为3，灰度图为1
num_classes：1000，表示有1000个目标类
depths：[2, 2, 6, 2]，模型深度，长度为4，表示有4个BasicLayer，每个BasicLayer的depth参数分别为2, 2, 6, 2
num_heads：[3, 6, 12, 24]，模型深度，长度为4，表示有4个BasicLayer，每个BasicLayer的num_heads参数分别为3, 6, 12, 24
以上是主要参数，其他参数大家自己理解，下面的部分介绍每个模块时，也会提到上面的其他参数。
2.transformer模型各模块的介绍
2.1 trunc_normal模块
trunc_normal是截断正态分布，用来初始化权重张量（用截断正态分布填充张量，a是下限，b是上限，mean是均值,正态分布的标准差,tensor要初始化的张量）。
代码如下：
import torch
import math
# 1.截断正态分布，用来初始化权重张量（用截断正态分布填充张量，a是下限，b是上限，mean是均值,正态分布的标准差,tensor要初始化的张量）
def trunc_normal_(tensor, mean=0., std=1., a=-2., b=2.):
def _no_grad_trunc_normal_(tensor, mean, std, a, b):
def norm_cdf(x):
return (1. + math.erf(x / math.sqrt(2.))) / 2.
with torch.no_grad():
l = norm_cdf((a - mean) / std)
u = norm_cdf((b - mean) / std)
tensor.uniform_(2 * l - 1, 2 * u - 1)
tensor.erfinv_()
tensor.mul_(std * math.sqrt(2.))
tensor.add_(mean)
tensor.clamp_(min=a, max=b)
return tensor
return _no_grad_trunc_normal_(tensor, mean, std, a, b)
if __name__ == "__main__":
tensor = torch.empty(3, 3)
outs = trunc_normal_(tensor, mean=0.0, std=1.0, a=-2.0, b=2.0)
print(outs)运行结果如下：
显然，该模块建立了一个3X3的tensor张量，且填充的数据符合均值为0，方差为1，最大值为2，最小值为-2的正态分布。
2.2 PatchEmbed模块
该模块是对输入进来的图片进行高和宽的压缩。
代码如下：
import torch
import torch.nn as nn
# 2.对输入进来的图片进行高和宽的压缩
class PatchEmbed(nn.Module):
def __init__(self, img_size=[224, 224], patch_size=4, in_chans=3, embed_dim=96, norm_layer=None):
super().__init__()
# [224, 224]
self.img_size = img_size
# [4, 4]
self.patch_size = [patch_size, patch_size]
# [56, 56]
self.patches_resolution = [self.img_size[0] // self.patch_size[0], self.img_size[1] // self.patch_size[1]]
# 3
self.in_chans = in_chans
# 96
self.embed_dim = embed_dim
# 3136
self.num_patches = self.patches_resolution[0] * self.patches_resolution[1]
# -------------------------------------------------------#
# bs, 224, 224, 3 -> bs, 56, 56, 96
# -------------------------------------------------------#
self.proj = nn.Conv2d(in_chans,embed_dim,kernel_size=patch_size, stride=patch_size)# 卷积结构
if norm_layer is not None:
self.norm = norm_layer(embed_dim)
else:
self.norm = None
def forward(self,x):
# x = self.proj(x) # bs, 3, 224, 224 -> bs, 96, 56, 56
# x = x.flatten(2) # bs, 96, 56, 56 -> 4, 96, 3136
# x = x.transpose(1, 2) # 4, 96, 3136 -> 4, 3136, 96
x = self.proj(x).flatten(2).transpose(1, 2)
if self.norm is not None:
x = self.norm(x)
return x
if __name__ == "__main__":
datas = torch.randn(4,3,224,224)
nets = PatchEmbed()
outs = nets(datas)
print(outs.size())运行结果如下：
压缩过程中数据大小的变化在代码的每一步注释的很清楚。
2.3 make_divisible模块
确保网络层的通道数可以被divisor（通常是8）整除： v:原始通道数； divisor：除数，通常是8； min_value：通道数的最小值。
代码如下：
# 3. 确保网络层的通道数可以被divisor（通常是8）整除： v:原始通道数； divisor：除数，通常是8； min_value：通道数的最小值。
def _make_divisible(v, divisor, min_value=None):
if min_value is None:
min_value = divisor
new_v = max(min_value, int(v + divisor / 2) // divisor * divisor)
if new_v < 0.9 * v:
new_v += divisor
return new_v
if __name__ == "__main__":
v = 10
divisor = 8
min_value = None
datas = _make_divisible(v, divisor, min_value=None)
print(datas)
运行结果如下：
2.4 window_partition模块
将图像分割成小块以便进行特征提取和处理.
代码如下：
import torch
# 4. 将图像分割成小块以便进行特征提取和处理
def window_partition(x, window_size):
B, H, W, C = x.shape
# ------------------------------------------------------------------#
# bs, 56, 56, 96 -> bs, 8, 7, 8, 7, 96 -> bs * 64, 7, 7, 96
# ------------------------------------------------------------------#
x = x.view(B, H // window_size, window_size, W // window_size, window_size, C)# bs, 8, 7, 8, 7, 96
#print(x.permute(0, 1, 3, 2, 4, 5).shape) ====> # bs, 8, 8, 7, 7, 96
windows = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(-1, window_size, window_size, C)
return windows
if __name__ == "__main__":
x = torch.randn(4, 56, 56, 96)
window_size = 7
windows = window_partition(x, window_size)
print(windows.shape)运行结果如下：
2.5 PatchMerging模块
对输入进来的特征层进行高和宽的压缩(降采样).
代码如下：
import torch
import torch.nn as nn
import math
import numpy as np
# 6.对输入进来的特征层进行高和宽的压缩(降采样)
class PatchMerging(nn.Module):
def __init__(self, input_resolution, dim, norm_layer=nn.LayerNorm):
super().__init__()
self.input_resolution = input_resolution
self.dim = dim
self.norm = norm_layer(4 * dim)
self.reduction = nn.Linear(4 * dim, 2 * dim, bias=False)
def forward(self, x):
H, W = self.input_resolution
B, L, C = x.shape
assert L == H * W, "input feature has wrong size"
assert H % 2 == 0 and W % 2 == 0, f"x size ({H}*{W}) are not even."
# -------------------------------------------------------#
# bs, 3136, 96 -> bs, 56, 56, 96
# -------------------------------------------------------#
x = x.view(B, H, W, C)
print("2.===> ",x.shape)
# -------------------------------------------------------#
# x0 ~ x3 bs, 56, 56, 96 -> bs, 28, 28, 96
# -------------------------------------------------------#
x0 = x[:, 0::2, 0::2, :] # B H/2 W/2 C
x1 = x[:, 1::2, 0::2, :] # B H/2 W/2 C
x2 = x[:, 0::2, 1::2, :] # B H/2 W/2 C
x3 = x[:, 1::2, 1::2, :] # B H/2 W/2 C
# -------------------------------------------------------#
# 4 X bs, 28, 28, 96 -> bs, 28, 28, 384
# -------------------------------------------------------#
x = torch.cat([x0, x1, x2, x3], -1) # B H/2 W/2 4*C
# -------------------------------------------------------#
# bs, 28, 28, 384 -> bs, 784, 384 # 通道数增加到4倍（96-->384）,value降低到1/4（3136-->784）
# -------------------------------------------------------#
x = x.view(B, -1, 4 * C) # B H/2*W/2 4*C
# -------------------------------------------------------#
# bs, 784, 384 -> bs, 784, 192
# -------------------------------------------------------#
print("3.===> ",x.shape)
x = self.norm(x)
print("4.====> ", x.shape)
x = self.reduction(x)
print("5.====> ", x.shape)
return x
if __name__ == "__main__":
x = torch.randn(4, 3136, 96)
input_resolution = [56,56]
dim = 96
x = x.to(torch.float32)
model = PatchMerging(input_resolution,dim)
print('--------------------------')
print('1.===> ', x.shape)
y = model(x)
运行结果如下：
2.6 WindowAttention模块
多头注意力机制：将序列划分为多个窗口，每个窗口内的元素进行注意力计算，从而减少计算量和显存需求.
代码如下：
import torch
import torch.nn as nn
import math
# 1.截断正态分布，用来初始化权重张量（用截断正态分布填充张量，a是下限，b是上限，mean是均值,正态分布的标准差,tensor要初始化的张量）
def trunc_normal_(tensor, mean=0., std=1., a=-2., b=2.):
def _no_grad_trunc_normal_(tensor, mean, std, a, b):
def norm_cdf(x):
return (1. + math.erf(x / math.sqrt(2.))) / 2.
with torch.no_grad():
l = norm_cdf((a - mean) / std)
u = norm_cdf((b - mean) / std)
tensor.uniform_(2 * l - 1, 2 * u - 1)
tensor.erfinv_()
tensor.mul_(std * math.sqrt(2.))
tensor.add_(mean)
tensor.clamp_(min=a, max=b)
return tensor
return _no_grad_trunc_normal_(tensor, mean, std, a, b)
# 7.多头注意力机制：将序列划分为多个窗口，每个窗口内的元素进行注意力计算，从而减少计算量和显存需求
class WindowAttention(nn.Module):
def __init__(self, dim, window_size, num_heads, qkv_bias=True, qk_scale=None, attn_drop=0., proj_drop=0.):
super().__init__()
self.dim = dim
self.window_size = window_size # Wh, Ww
self.num_heads = num_heads
head_dim = dim // num_heads
self.scale = qk_scale or head_dim ** -0.5
# --------------------------------------------------------------------------#
# 相对坐标矩阵，用于表示每个窗口内，其它点相对于自己的坐标
# 由于相对坐标取值范围为-6 ~ +6。中间共13个值，因此需要13 * 13
# 13 * 13, num_heads
# --------------------------------------------------------------------------#
self.relative_position_bias_table = nn.Parameter(torch.zeros((2 * window_size[0] - 1) * (2 * window_size[1] - 1), num_heads))
# --------------------------------------------------------------------------#
# 该部分用于获取7x7的矩阵内部，其它特征点相对于自身相对坐标
# --------------------------------------------------------------------------#
coords_h = torch.arange(self.window_size[0])
coords_w = torch.arange(self.window_size[1])
coords = torch.stack(torch.meshgrid([coords_h, coords_w])) # 2, Wh, Ww
coords_flatten = torch.flatten(coords, 1) # 2, Wh*Ww
relative_coords = coords_flatten[:, :, None] - coords_flatten[:, None, :] # 2, Wh*Ww, Wh*Ww
relative_coords = relative_coords.permute(1, 2, 0).contiguous() # Wh*Ww, Wh*Ww, 2
relative_coords[:, :, 0] += self.window_size[0] - 1 # shift to start from 0
relative_coords[:, :, 1] += self.window_size[1] - 1
relative_coords[:, :, 0] *= 2 * self.window_size[1] - 1
relative_position_index = relative_coords.sum(-1) # Wh*Ww, Wh*Ww
self.register_buffer("relative_position_index", relative_position_index)
# --------------------------------------------------------------------------#
# 乘积获得q、k、v，用于计算多头注意力机制
# --------------------------------------------------------------------------#
self.qkv = nn.Linear(dim, dim * 3, bias=qkv_bias)# 通道数（线性变换）： dim ===> 3*dim
self.attn_drop = nn.Dropout(attn_drop)# 防止过拟合
self.proj = nn.Linear(dim, dim)# 通道数（线性变换）： dim ===> dim
self.proj_drop = nn.Dropout(proj_drop)# 防止过拟合
trunc_normal_(self.relative_position_bias_table, std=.02)
self.softmax = nn.Softmax(dim=-1)
def forward(self, x, mask=None):
B_, N, C = x.shape
print("1.====> ",B_, N, C)
# --------------------------------------------------------------------------#
# bs * 64, 49, 96 -> bs * 64, 49, 96 * 3 ->
# bs * 64, 49, 3, num_heads, 32 -> 3, bs * 64, num_head, 49, 32
# --------------------------------------------------------------------------#
qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4)
print("2.====> ", qkv.size())
# --------------------------------------------------------------------------#
# bs * 64, num_head, 49, 32
# --------------------------------------------------------------------------#
q, k, v = qkv[0], qkv[1], qkv[2]
print("3.====> ", q.size(), k.size(), v.size())
# --------------------------------------------------------------------------#
# bs * 64, num_head, 49, 49
# --------------------------------------------------------------------------#
q = q * self.scale
attn = (q @ k.transpose(-2, -1)) # 矩阵的点乘：[49,12] @ [12, 49] = [49, 49]
print("4.====> ", attn.size())
# --------------------------------------------------------------------------#
# 这一步是根据已经求得的注意力，加上相对坐标的偏执量
# 形成最后的注意力
# --------------------------------------------------------------------------#
relative_position_bias = self.relative_position_bias_table[self.relative_position_index.view(-1)].view(
self.window_size[0] * self.window_size[1], self.window_size[0] * self.window_size[1], -1) # Wh*Ww,Wh*Ww,nH
print("5.====> ", relative_position_bias.size())
relative_position_bias = relative_position_bias.permute(2, 0, 1).contiguous() # nH, Wh*Ww,Wh*Ww
attn = attn + relative_position_bias.unsqueeze(0)
print("6.====> ", attn.size())
# --------------------------------------------------------------------------#
# 加上mask，保证分区。
# bs * 64, num_head, 49, 49
# --------------------------------------------------------------------------#
if mask is not None:
nW = mask.shape[0]
attn = attn.view(B_ // nW, nW, self.num_heads, N, N) + mask.unsqueeze(1).unsqueeze(0)
attn = attn.view(-1, self.num_heads, N, N)
attn = self.softmax(attn)
else:
attn = self.softmax(attn)
attn = self.attn_drop(attn)
print("7.====> ", attn.size())
# ---------------------------------------------------------------------------------------#
# bs * 64, num_head, 49, 49 @ bs * 64, num_head, 49, 32 -> bs * 64, num_head, 49, 32
#
# bs * 64, num_head, 49, 32 -> bs * 64, 49, 96
# ---------------------------------------------------------------------------------------#
x = (attn @ v).transpose(1, 2).reshape(B_, N, C)
print("8.====> ", x.size())
x = self.proj(x)
print("9.====> ", x.size())
x = self.proj_drop(x)
print("10.====> ", x.size())
return x
if __name__ == "__main__":
# x = torch.randn(4, 3136, 96)
x = torch.randn(256, 49, 96)
dim = 96
window_size = [7,7]
num_heads = 8
model = WindowAttention(dim, window_size, num_heads, qkv_bias=True, qk_scale=None, attn_drop=0., proj_drop=0.)
outs = model(x)
运行结果如下：
2.7 drop_path模块
生成一系列的mask来选择网络中的分支。mask为1的地方，保留相应的网络结构；mask为0的地方，使该部分网络结构失效
代码如下：
import torch.nn as nn
import torch
# 8.生成一系列的mask来选择网络中的分支。mask为1的地方，保留相应的网络结构；mask为0的地方，使该部分网络结构失效
def drop_path(x, drop_prob: float = 0., training: bool = False, scale_by_keep: bool = True):
"""
Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).
This is the same as the DropConnect impl I created for EfficientNet, etc networks, however,
the original name is misleading as 'Drop Connect' is a different form of dropout in a separate paper...
See discussion: https://github.com/tensorflow/tpu/issues/494#issuecomment-532968956 ... I've opted for
changing the layer and argument names to 'drop path' rather than mix DropConnect as a layer name and use
'survival rate' as the argument.
"""
if drop_prob == 0. or not training:
return x
keep_prob = 1 - drop_prob
shape = (x.shape[0],) + (1,) * (x.ndim - 1) # work with diff dim tensors, not just 2D ConvNets
random_tensor = x.new_empty(shape).bernoulli_(keep_prob)
if keep_prob > 0.0 and scale_by_keep:
random_tensor.div_(keep_prob)
return x * random_tensor
class DropPath(nn.Module):
"""
Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).
"""
def __init__(self, drop_prob=None, scale_by_keep=True):
super(DropPath, self).__init__()
self.drop_prob = drop_prob
self.scale_by_keep = scale_by_keep
def forward(self, x):
return drop_path(x, self.drop_prob, self.training, self.scale_by_keep)
if __name__ == "__main__":
# x = torch.randn(4, 3136, 96)
x = torch.randn(5, 3, 3)
nets =DropPath(drop_prob = 0.5)
outs = nets(x)
print(outs)
运行结果如下：
2.8 Mlp模块
该模块是进行两次全连接。
代码如下：
import torch.nn as nn
import torch
import numpy as np
# 5.激活函数GELU
class GELU(nn.Module):
def __init__(self):
super(GELU, self).__init__()
def forward(self,x):
return 0.5 * x * (1 + torch.tanh(np.sqrt(2 / np.pi) * (x + 0.044715 * torch.pow(x, 3))))
# 9. 两次全连接
class Mlp(nn.Module):
def __init__(self, in_features, hidden_features=None, out_features=None, act_layer=GELU, drop=0.):
super().__init__()
out_features = out_features or in_features
hidden_features = hidden_features or in_features
self.fc1 = nn.Linear(in_features, hidden_features)
self.act = act_layer()
self.fc2 = nn.Linear(hidden_features, out_features)
self.drop = nn.Dropout(drop)
def forward(self, x):
x = self.fc1(x)
x = self.act(x)
x = self.drop(x)
print("1.====> ", x.size())
x = self.fc2(x)
x = self.drop(x)
print("2.====> ", x.size())
return x
if __name__ == "__main__":
#x = torch.randn(4, 3136, 96)
x = torch.randn(2, 2, 4, 4)
nets =Mlp(4, 8,10)
outs = nets(x)
# print("3.====> ",outs)
运行结果如下：
2.9 window_reverse模块
该模块是将窗口内的信息重新组合回原始的特征图或图像。
代码如下：
import torch
# 10.将窗口内的信息重新组合回原始的特征图或图像
def window_reverse(windows, window_size, H, W):
# ------------------------------------------------------------------#
# bs * 64, 7, 7, 96 -> bs, 8, 8, 7, 7, 96 -> bs, 56, 56, 96
# ------------------------------------------------------------------#
B = int(windows.shape[0] / (H * W / window_size / window_size))
x = windows.view(B, H // window_size, W // window_size, window_size, window_size, -1)
x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1)
return x
if __name__ == "__main__":
#x = torch.randn(4, 3136, 96)
windows = torch.randn(256, 7, 7, 96)
window_size = 7
H = 56
W = 56
outs = window_reverse(windows, window_size, H, W)
print(outs.size())
运行结果如下：
2.10 SwinTransformerBlock模块
该模块是transfomer的每层的基础模块，具体解释在代码中有详细的解释。
代码如下：
import math
import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.utils.checkpoint as checkpoint
# 1.截断正态分布，用来初始化权重张量（用截断正态分布填充张量，a是下限，b是上限，mean是均值,正态分布的标准差,tensor要初始化的张量）
def trunc_normal_(tensor, mean=0., std=1., a=-2., b=2.):
def _no_grad_trunc_normal_(tensor, mean, std, a, b):
def norm_cdf(x):
return (1. + math.erf(x / math.sqrt(2.))) / 2.
with torch.no_grad():
l = norm_cdf((a - mean) / std)
u = norm_cdf((b - mean) / std)
tensor.uniform_(2 * l - 1, 2 * u - 1)
tensor.erfinv_()
tensor.mul_(std * math.sqrt(2.))
tensor.add_(mean)
tensor.clamp_(min=a, max=b)
return tensor
return _no_grad_trunc_normal_(tensor, mean, std, a, b)
# 3. 确保网络层的通道数可以被divisor（通常是8）整除： v:原始通道数； divisor：除数，通常是8； min_value：通道数的最小值。
def _make_divisible(v, divisor, min_value=None):
if min_value is None:
min_value = divisor
new_v = max(min_value, int(v + divisor / 2) // divisor * divisor)
if new_v < 0.9 * v:
new_v += divisor
return new_v
# 4. 将图像分割成小块以便进行特征提取和处理
def window_partition(x, window_size):
B, H, W, C = x.shape
# ------------------------------------------------------------------#
# bs, 56, 56, 96 -> bs, 8, 7, 8, 7, 96 -> bs * 64, 7, 7, 96
# ------------------------------------------------------------------#
x = x.view(B, H // window_size, window_size, W // window_size, window_size, C)
windows = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(-1, window_size, window_size, C)
return windows
# 5.激活函数GELU
class GELU(nn.Module):
def __init__(self):
super(GELU, self).__init__()
def forward(self,x):
return 0.5 * x * (1 + torch.tanh(np.sqrt(2 / np.pi) * (x + 0.044715 * torch.pow(x, 3))))
# 7.注意力机制：将序列划分为多个窗口，每个窗口内的元素进行注意力计算，从而减少计算量和显存需求
class WindowAttention(nn.Module):
def __init__(self, dim, window_size, num_heads, qkv_bias=True, qk_scale=None, attn_drop=0., proj_drop=0.):
super().__init__()
self.dim = dim
self.window_size = window_size # Wh, Ww
self.num_heads = num_heads
head_dim = dim // num_heads
self.scale = qk_scale or head_dim ** -0.5
# --------------------------------------------------------------------------#
# 相对坐标矩阵，用于表示每个窗口内，其它点相对于自己的坐标
# 由于相对坐标取值范围为-6 ~ +6。中间共13个值，因此需要13 * 13
# 13 * 13, num_heads
# --------------------------------------------------------------------------#
self.relative_position_bias_table = nn.Parameter(
torch.zeros((2 * window_size[0] - 1) * (2 * window_size[1] - 1), num_heads)
)
# --------------------------------------------------------------------------#
# 该部分用于获取7x7的矩阵内部，其它特征点相对于自身相对坐标
# --------------------------------------------------------------------------#
coords_h = torch.arange(self.window_size[0])
coords_w = torch.arange(self.window_size[1])
coords = torch.stack(torch.meshgrid([coords_h, coords_w])) # 2, Wh, Ww
coords_flatten = torch.flatten(coords, 1) # 2, Wh*Ww
relative_coords = coords_flatten[:, :, None] - coords_flatten[:, None, :] # 2, Wh*Ww, Wh*Ww
relative_coords = relative_coords.permute(1, 2, 0).contiguous() # Wh*Ww, Wh*Ww, 2
relative_coords[:, :, 0] += self.window_size[0] - 1 # shift to start from 0
relative_coords[:, :, 1] += self.window_size[1] - 1
relative_coords[:, :, 0] *= 2 * self.window_size[1] - 1
relative_position_index = relative_coords.sum(-1) # Wh*Ww, Wh*Ww
self.register_buffer("relative_position_index", relative_position_index)
# --------------------------------------------------------------------------#
# 乘积获得q、k、v，用于计算多头注意力机制
# --------------------------------------------------------------------------#
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)
trunc_normal_(self.relative_position_bias_table, std=.02)
self.softmax = nn.Softmax(dim=-1)
def forward(self, x, mask=None):
B_, N, C = x.shape
# --------------------------------------------------------------------------#
# bs * 64, 49, 96 -> bs * 64, 49, 96 * 3 ->
# bs * 64, 49, 3, num_heads, 32 -> 3, bs * 64, num_head, 49, 32
# --------------------------------------------------------------------------#
qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4)
# --------------------------------------------------------------------------#
# bs * 64, num_head, 49, 32
# --------------------------------------------------------------------------#
q, k, v = qkv[0], qkv[1], qkv[2]
# --------------------------------------------------------------------------#
# bs * 64, num_head, 49, 49
# --------------------------------------------------------------------------#
q = q * self.scale
attn = (q @ k.transpose(-2, -1))
# --------------------------------------------------------------------------#
# 这一步是根据已经求得的注意力，加上相对坐标的偏执量
# 形成最后的注意力
# --------------------------------------------------------------------------#
relative_position_bias = self.relative_position_bias_table[self.relative_position_index.view(-1)].view(
self.window_size[0] * self.window_size[1], self.window_size[0] * self.window_size[1], -1) # Wh*Ww,Wh*Ww,nH
relative_position_bias = relative_position_bias.permute(2, 0, 1).contiguous()
attn = attn + relative_position_bias.unsqueeze(0)
# --------------------------------------------------------------------------#
# 加上mask，保证分区。
# bs * 64, num_head, 49, 49
# --------------------------------------------------------------------------#
if mask is not None:
nW = mask.shape[0]
attn = attn.view(B_ // nW, nW, self.num_heads, N, N) + mask.unsqueeze(1).unsqueeze(0)
attn = attn.view(-1, self.num_heads, N, N)
attn = self.softmax(attn)
else:
attn = self.softmax(attn)
attn = self.attn_drop(attn)
# ---------------------------------------------------------------------------------------#
# bs * 64, num_head, 49, 49 @ bs * 64, num_head, 49, 32 -> bs * 64, num_head, 49, 32
#
# bs * 64, num_head, 49, 32 -> bs * 64, 49, 96
# ---------------------------------------------------------------------------------------#
x = (attn @ v).transpose(1, 2).reshape(B_, N, C)
x = self.proj(x)
x = self.proj_drop(x)
return x
# 8.生成一系列的mask来选择网络中的分支。mask为1的地方，保留相应的网络结构；mask为0的地方，使该部分网络结构失效
def drop_path(x, drop_prob: float = 0., training: bool = False, scale_by_keep: bool = True):
"""
Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).
This is the same as the DropConnect impl I created for EfficientNet, etc networks, however,
the original name is misleading as 'Drop Connect' is a different form of dropout in a separate paper...
See discussion: https://github.com/tensorflow/tpu/issues/494#issuecomment-532968956 ... I've opted for
changing the layer and argument names to 'drop path' rather than mix DropConnect as a layer name and use
'survival rate' as the argument.
"""
if drop_prob == 0. or not training:
return x
keep_prob = 1 - drop_prob
shape = (x.shape[0],) + (1,) * (x.ndim - 1) # work with diff dim tensors, not just 2D ConvNets
random_tensor = x.new_empty(shape).bernoulli_(keep_prob)
if keep_prob > 0.0 and scale_by_keep:
random_tensor.div_(keep_prob)
return x * random_tensor
class DropPath(nn.Module):
"""
Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).
"""
def __init__(self, drop_prob=None, scale_by_keep=True):
super(DropPath, self).__init__()
self.drop_prob = drop_prob
self.scale_by_keep = scale_by_keep
def forward(self, x):
return drop_path(x, self.drop_prob, self.training, self.scale_by_keep)
# 9. 两次全连接
class Mlp(nn.Module):
def __init__(self, in_features, hidden_features=None, out_features=None, act_layer=GELU, drop=0.):
super().__init__()
out_features = out_features or in_features
hidden_features = hidden_features or in_features
self.fc1 = nn.Linear(in_features, hidden_features)
self.act = act_layer()
self.fc2 = nn.Linear(hidden_features, out_features)
self.drop = nn.Dropout(drop)
def forward(self, x):
x = self.fc1(x)
x = self.act(x)
x = self.drop(x)
x = self.fc2(x)
x = self.drop(x)
return x
# 10.将窗口内的信息重新组合回原始的特征图或图像
def window_reverse(windows, window_size, H, W):
# ------------------------------------------------------------------#
# bs * 64, 7, 7, 96 -> bs, 8, 8, 7, 7, 96 -> bs, 56, 56, 96
# ------------------------------------------------------------------#
B = int(windows.shape[0] / (H * W / window_size / window_size))
x = windows.view(B, H // window_size, W // window_size, window_size, window_size, -1)
x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1)
return x
# 11.每个阶段重复的基础模块
# -------------------------------------------------------#
# 每个阶段重复的基础模块
# 在这其中会使用WindowAttention进行特征提取
# -------------------------------------------------------#
class SwinTransformerBlock(nn.Module):
def __init__(self, dim, input_resolution, num_heads, window_size=7, shift_size=0,
mlp_ratio=4., qkv_bias=True, qk_scale=None, drop=0., attn_drop=0., drop_path=0.,
act_layer=GELU, norm_layer=nn.LayerNorm):
super().__init__()
self.dim = dim
self.input_resolution = input_resolution
self.num_heads = num_heads
self.window_size = window_size
self.shift_size = shift_size
self.mlp_ratio = mlp_ratio
if min(self.input_resolution) <= self.window_size:
self.shift_size = 0
self.window_size = min(self.input_resolution)
assert 0 <= self.shift_size < self.window_size, "shift_size must in 0-window_size"
self.norm1 = norm_layer(dim) # 正态化
self.attn = WindowAttention( # 多头注意力机制：将序列划分为多个窗口，每个窗口内的元素进行注意力计算，从而减少计算量和显存需求
dim,
window_size=[self.window_size, self.window_size],
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)
mlp_hidden_dim = int(dim * mlp_ratio)
self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop)
if self.shift_size > 0:
# ----------------------------------------------------------------#
# 由于进行特征提取时，会对输入的特征层进行的平移
# 如：
# [ [
# [1, 2, 3], [5, 6, 4],
# [4, 5, 6], --> [8, 9, 7],
# [7, 8, 9], [1, 2, 3],
# ] ]
# 这一步的作用就是使得平移后的区域块只计算自己部分的注意力机制
# ----------------------------------------------------------------#
H, W = self.input_resolution
_H, _W = _make_divisible(H, self.window_size), _make_divisible(W, self.window_size),
img_mask = torch.zeros((1, _H, _W, 1)) # 1 H W 1
h_slices = (slice(0, -self.window_size),
slice(-self.window_size, -self.shift_size),
slice(-self.shift_size, None))
w_slices = (slice(0, -self.window_size),
slice(-self.window_size, -self.shift_size),
slice(-self.shift_size, None))
cnt = 0
for h in h_slices:
for w in w_slices:
img_mask[:, h, w, :] = cnt
cnt += 1
mask_windows = window_partition(img_mask, self.window_size) # nW, window_size, window_size, 1
mask_windows = mask_windows.view(-1, self.window_size * self.window_size)
attn_mask = mask_windows.unsqueeze(1) - mask_windows.unsqueeze(2)
attn_mask = attn_mask.masked_fill(attn_mask != 0, float(-100.0)).masked_fill(attn_mask == 0, float(0.0))
self.attn_mask = attn_mask.cpu().numpy()
else:
self.attn_mask = None
def forward(self, x):
H, W = self.input_resolution
B, L, C = x.shape
assert L == H * W, "input feature has wrong size"
# -----------------------------------------------#
# bs, 3136, 96 -> bs, 56, 56, 96
# -----------------------------------------------#
shortcut = x
x = self.norm1(x) # [4, 3136, 96]
print("1.====> ", x.size())
x = x.view(B, H, W, C) # [4, 56, 56, 96]
print("2.====> ", x.size())
_H, _W = _make_divisible(H, self.window_size), _make_divisible(W, self.window_size), # 56 56
print("3.====> ", _H, _W )
x = x.permute(0, 3, 1, 2) # [4, 96, 56, 56]
print("4.====> ", x.size())
x = F.interpolate(x, [_H, _W], mode='bicubic', align_corners=False).permute(0, 2, 3, 1) # 插值计算 [4, 56, 56, 96]
print("5.====> ", x.size())
# -----------------------------------------------#
# 进行特征层的平移
# -----------------------------------------------#
if self.shift_size > 0:
shifted_x = torch.roll(x, shifts=(-self.shift_size, -self.shift_size), dims=(1, 2))
else:
shifted_x = x # [4, 56, 56, 96]
print("6.====> ", shifted_x.size())
# ------------------------------------------------------------------------------------------#
# bs, 56, 56, 96 -> bs * 64, 7, 7, 96 -> bs * 64, 49, 96
# ------------------------------------------------------------------------------------------#
x_windows = window_partition(shifted_x, self.window_size) # num_windows * B, window_size, window_size, C
x_windows = x_windows.view(-1, self.window_size * self.window_size, C) # nW*B, window_size*window_size, C <====> [256, 49, 96]
print("7.====> ", x_windows.size())
# -----------------------------------------------#
# bs * 64, 49, 97 -> bs * 64, 49, 97
# -----------------------------------------------#
if type(self.attn_mask) != type(None):
attn_mask = torch.tensor(self.attn_mask).cuda() if x.is_cuda else torch.tensor(self.attn_mask)
else:
attn_mask = None
attn_windows = self.attn(x_windows, mask=attn_mask) # nW*B, window_size*window_size, C <====> [256, 49, 96]
print("8.====> ", attn_windows.size())
# -----------------------------------------------#
# bs * 64, 49, 97 -> bs, 56, 56, 96
# -----------------------------------------------#
attn_windows = attn_windows.view(-1, self.window_size, self.window_size, C) # [256, 7, 7, 96]
print("9.====> ", attn_windows.size())
shifted_x = window_reverse(attn_windows, self.window_size, _H, _W) # B H' W' C <=====> [4, 56, 56, 96]
print("10.====> ", shifted_x.size())
# -----------------------------------------------#
# 将特征层平移回来
# -----------------------------------------------#
if self.shift_size > 0:
x = torch.roll(shifted_x, shifts=(self.shift_size, self.shift_size), dims=(1, 2))
else:
x = shifted_x # [4, 56, 56, 96]
print("11.====> ", x.size())
x = x.permute(0, 3, 1, 2) # [4, 96, 56, 56]
print("12.====> ", x.size())
x = F.interpolate(x, [H, W], mode='bicubic', align_corners=False).permute(0, 2, 3, 1) # 插值法计算 [4, 56, 56, 96]
print("13.====> ", x.size())
# -----------------------------------------------#
# bs, 3136, 96
# -----------------------------------------------#
x = x.view(B, H * W, C) # [4, 3136, 96]
print("14.====> ", x.size())
# -----------------------------------------------#
# FFN
# bs, 3136, 96
# -----------------------------------------------#
x = shortcut + self.drop_path(x)
x = x + self.drop_path(self.mlp(self.norm2(x))) # [4, 3136, 96]
print("15.====> ", x.size())
return x
if __name__ == "__main__":
x = torch.randn(4, 3136, 96)
net = SwinTransformerBlock(dim = 96, input_resolution = [56,56], num_heads = 8)
outs = net(x)
# print("10.====> ",outs.size())
运行结果如下：
2.11 BasicLayer模块
该模块是Swin-Transformer的基础模块:其实就是对 SwinTransformerBlock 模块取不同的参数，然后加一个下采样的选择。
代码如下：
import math
import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.utils.checkpoint as checkpoint
# 1.截断正态分布，用来初始化权重张量（用截断正态分布填充张量，a是下限，b是上限，mean是均值,正态分布的标准差,tensor要初始化的张量）
def trunc_normal_(tensor, mean=0., std=1., a=-2., b=2.):
def _no_grad_trunc_normal_(tensor, mean, std, a, b):
def norm_cdf(x):
return (1. + math.erf(x / math.sqrt(2.))) / 2.
with torch.no_grad():
l = norm_cdf((a - mean) / std)
u = norm_cdf((b - mean) / std)
tensor.uniform_(2 * l - 1, 2 * u - 1)
tensor.erfinv_()
tensor.mul_(std * math.sqrt(2.))
tensor.add_(mean)
tensor.clamp_(min=a, max=b)
return tensor
return _no_grad_trunc_normal_(tensor, mean, std, a, b)
# 2.对输入进来的图片进行高和宽的压缩
class PatchEmbed(nn.Module):
def __init__(self, img_size=[224, 224], patch_size=4, in_chans=3, embed_dim=96, norm_layer=None):
super().__init__()
# [224, 224]
self.img_size = img_size
# [4, 4]
self.patch_size = [patch_size, patch_size]
# [56, 56]
self.patches_resolution = [self.img_size[0] // self.patch_size[0], self.img_size[1] // self.patch_size[1]]
# 3
self.in_chans = in_chans
# 96
self.embed_dim = embed_dim
# 3136
self.num_patches = self.patches_resolution[0] * self.patches_resolution[1]
# -------------------------------------------------------#
# bs, 224, 224, 3 -> bs, 56, 56, 96
# -------------------------------------------------------#
self.proj = nn.Conv2d(in_chans,embed_dim,kernel_size=patch_size, stride=patch_size)# 卷积结构
if norm_layer is not None:
self.norm = norm_layer(embed_dim)
else:
self.norm = None
def forward(self,x):
# x = self.proj(x) # bs, 3, 224, 224 -> bs, 96, 56, 56
# x = x.flatten(2) # bs, 96, 56, 56 -> 4, 96, 3136
# x = x.transpose(1, 2) # 4, 96, 3136 -> 4, 3136, 96
x = self.proj(x).flatten(2).transpose(1, 2)
if self.norm is not None:
x = self.norm(x)
return x
# 3. 确保网络层的通道数可以被divisor（通常是8）整除： v:原始通道数； divisor：除数，通常是8； min_value：通道数的最小值。
def _make_divisible(v, divisor, min_value=None):
if min_value is None:
min_value = divisor
new_v = max(min_value, int(v + divisor / 2) // divisor * divisor)
if new_v < 0.9 * v:
new_v += divisor
return new_v
# 4. 将图像分割成小块以便进行特征提取和处理
def window_partition(x, window_size):
B, H, W, C = x.shape
# ------------------------------------------------------------------#
# bs, 56, 56, 96 -> bs, 8, 7, 8, 7, 96 -> bs * 64, 7, 7, 96
# ------------------------------------------------------------------#
x = x.view(B, H // window_size, window_size, W // window_size, window_size, C)
windows = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(-1, window_size, window_size, C)
return windows
# 5.激活函数GELU
class GELU(nn.Module):
def __init__(self):
super(GELU, self).__init__()
def forward(self,x):
return 0.5 * x * (1 + torch.tanh(np.sqrt(2 / np.pi) * (x + 0.044715 * torch.pow(x, 3))))
# 6.对输入进来的特征层进行高和宽的压缩
# -------------------------------------------------------#
# 对输入进来的特征层进行高和宽的压缩
# 进行跨特征点的特征提取，提取完成后进行堆叠。
# -------------------------------------------------------#
class PatchMerging(nn.Module):
def __init__(self, input_resolution, dim, norm_layer=nn.LayerNorm):
super().__init__()
self.input_resolution = input_resolution
self.dim = dim
self.norm = norm_layer(4 * dim)
self.reduction = nn.Linear(4 * dim, 2 * dim, bias=False)
def forward(self, x):
H, W = self.input_resolution
B, L, C = x.shape
assert L == H * W, "input feature has wrong size"
assert H % 2 == 0 and W % 2 == 0, f"x size ({H}*{W}) are not even."
# -------------------------------------------------------#
# bs, 3136, 96 -> bs, 56, 56, 96
# -------------------------------------------------------#
x = x.view(B, H, W, C)
# -------------------------------------------------------#
# x0 ~ x3 bs, 56, 56, 96 -> bs, 28, 28, 96
# -------------------------------------------------------#
x0 = x[:, 0::2, 0::2, :] # B H/2 W/2 C
x1 = x[:, 1::2, 0::2, :] # B H/2 W/2 C
x2 = x[:, 0::2, 1::2, :] # B H/2 W/2 C
x3 = x[:, 1::2, 1::2, :] # B H/2 W/2 C
# -------------------------------------------------------#
# 4 X bs, 28, 28, 96 -> bs, 28, 28, 384
# -------------------------------------------------------#
x = torch.cat([x0, x1, x2, x3], -1) # B H/2 W/2 4*C
# -------------------------------------------------------#
# bs, 28, 28, 384 -> bs, 784, 384
# -------------------------------------------------------#
x = x.view(B, -1, 4 * C) # B H/2*W/2 4*C
# -------------------------------------------------------#
# bs, 784, 384 -> bs, 784, 192
# -------------------------------------------------------#
x = self.norm(x)
x = self.reduction(x)
return x
# 7.注意力机制：将序列划分为多个窗口，每个窗口内的元素进行注意力计算，从而减少计算量和显存需求
class WindowAttention(nn.Module):
def __init__(self, dim, window_size, num_heads, qkv_bias=True, qk_scale=None, attn_drop=0., proj_drop=0.):
super().__init__()
self.dim = dim
self.window_size = window_size # Wh, Ww
self.num_heads = num_heads
head_dim = dim // num_heads
self.scale = qk_scale or head_dim ** -0.5
# --------------------------------------------------------------------------#
# 相对坐标矩阵，用于表示每个窗口内，其它点相对于自己的坐标
# 由于相对坐标取值范围为-6 ~ +6。中间共13个值，因此需要13 * 13
# 13 * 13, num_heads
# --------------------------------------------------------------------------#
self.relative_position_bias_table = nn.Parameter(
torch.zeros((2 * window_size[0] - 1) * (2 * window_size[1] - 1), num_heads)
)
# --------------------------------------------------------------------------#
# 该部分用于获取7x7的矩阵内部，其它特征点相对于自身相对坐标
# --------------------------------------------------------------------------#
coords_h = torch.arange(self.window_size[0])
coords_w = torch.arange(self.window_size[1])
coords = torch.stack(torch.meshgrid([coords_h, coords_w])) # 2, Wh, Ww
coords_flatten = torch.flatten(coords, 1) # 2, Wh*Ww
relative_coords = coords_flatten[:, :, None] - coords_flatten[:, None, :] # 2, Wh*Ww, Wh*Ww
relative_coords = relative_coords.permute(1, 2, 0).contiguous() # Wh*Ww, Wh*Ww, 2
relative_coords[:, :, 0] += self.window_size[0] - 1 # shift to start from 0
relative_coords[:, :, 1] += self.window_size[1] - 1
relative_coords[:, :, 0] *= 2 * self.window_size[1] - 1
relative_position_index = relative_coords.sum(-1) # Wh*Ww, Wh*Ww
self.register_buffer("relative_position_index", relative_position_index)
# --------------------------------------------------------------------------#
# 乘积获得q、k、v，用于计算多头注意力机制
# --------------------------------------------------------------------------#
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)
trunc_normal_(self.relative_position_bias_table, std=.02)
self.softmax = nn.Softmax(dim=-1)
def forward(self, x, mask=None):
B_, N, C = x.shape
# --------------------------------------------------------------------------#
# bs * 64, 49, 96 -> bs * 64, 49, 96 * 3 ->
# bs * 64, 49, 3, num_heads, 32 -> 3, bs * 64, num_head, 49, 32
# --------------------------------------------------------------------------#
qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4)
# --------------------------------------------------------------------------#
# bs * 64, num_head, 49, 32
# --------------------------------------------------------------------------#
q, k, v = qkv[0], qkv[1], qkv[2]
# --------------------------------------------------------------------------#
# bs * 64, num_head, 49, 49
# --------------------------------------------------------------------------#
q = q * self.scale
attn = (q @ k.transpose(-2, -1))
# --------------------------------------------------------------------------#
# 这一步是根据已经求得的注意力，加上相对坐标的偏执量
# 形成最后的注意力
# --------------------------------------------------------------------------#
relative_position_bias = self.relative_position_bias_table[self.relative_position_index.view(-1)].view(
self.window_size[0] * self.window_size[1], self.window_size[0] * self.window_size[1], -1) # Wh*Ww,Wh*Ww,nH
relative_position_bias = relative_position_bias.permute(2, 0, 1).contiguous()
attn = attn + relative_position_bias.unsqueeze(0)
# --------------------------------------------------------------------------#
# 加上mask，保证分区。
# bs * 64, num_head, 49, 49
# --------------------------------------------------------------------------#
if mask is not None:
nW = mask.shape[0]
attn = attn.view(B_ // nW, nW, self.num_heads, N, N) + mask.unsqueeze(1).unsqueeze(0)
attn = attn.view(-1, self.num_heads, N, N)
attn = self.softmax(attn)
else:
attn = self.softmax(attn)
attn = self.attn_drop(attn)
# ---------------------------------------------------------------------------------------#
# bs * 64, num_head, 49, 49 @ bs * 64, num_head, 49, 32 -> bs * 64, num_head, 49, 32
#
# bs * 64, num_head, 49, 32 -> bs * 64, 49, 96
# ---------------------------------------------------------------------------------------#
x = (attn @ v).transpose(1, 2).reshape(B_, N, C)
x = self.proj(x)
x = self.proj_drop(x)
return x
# 8.生成一系列的mask来选择网络中的分支。mask为1的地方，保留相应的网络结构；mask为0的地方，使该部分网络结构失效
def drop_path(x, drop_prob: float = 0., training: bool = False, scale_by_keep: bool = True):
"""
Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).
This is the same as the DropConnect impl I created for EfficientNet, etc networks, however,
the original name is misleading as 'Drop Connect' is a different form of dropout in a separate paper...
See discussion: https://github.com/tensorflow/tpu/issues/494#issuecomment-532968956 ... I've opted for
changing the layer and argument names to 'drop path' rather than mix DropConnect as a layer name and use
'survival rate' as the argument.
"""
if drop_prob == 0. or not training:
return x
keep_prob = 1 - drop_prob
shape = (x.shape[0],) + (1,) * (x.ndim - 1) # work with diff dim tensors, not just 2D ConvNets
random_tensor = x.new_empty(shape).bernoulli_(keep_prob)
if keep_prob > 0.0 and scale_by_keep:
random_tensor.div_(keep_prob)
return x * random_tensor
class DropPath(nn.Module):
"""
Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).
"""
def __init__(self, drop_prob=None, scale_by_keep=True):
super(DropPath, self).__init__()
self.drop_prob = drop_prob
self.scale_by_keep = scale_by_keep
def forward(self, x):
return drop_path(x, self.drop_prob, self.training, self.scale_by_keep)
# 9. 两次全连接
class Mlp(nn.Module):
def __init__(self, in_features, hidden_features=None, out_features=None, act_layer=GELU, drop=0.):
super().__init__()
out_features = out_features or in_features
hidden_features = hidden_features or in_features
self.fc1 = nn.Linear(in_features, hidden_features)
self.act = act_layer()
self.fc2 = nn.Linear(hidden_features, out_features)
self.drop = nn.Dropout(drop)
def forward(self, x):
x = self.fc1(x)
x = self.act(x)
x = self.drop(x)
x = self.fc2(x)
x = self.drop(x)
return x
# 10.将窗口内的信息重新组合回原始的特征图或图像
def window_reverse(windows, window_size, H, W):
# ------------------------------------------------------------------#
# bs * 64, 7, 7, 96 -> bs, 8, 8, 7, 7, 96 -> bs, 56, 56, 96
# ------------------------------------------------------------------#
B = int(windows.shape[0] / (H * W / window_size / window_size))
x = windows.view(B, H // window_size, W // window_size, window_size, window_size, -1)
x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1)
return x
# 11.每个阶段重复的基础模块
# -------------------------------------------------------#
# 每个阶段重复的基础模块
# 在这其中会使用WindowAttention进行特征提取
# -------------------------------------------------------#
class SwinTransformerBlock(nn.Module):
def __init__(self, dim, input_resolution, num_heads, window_size=7, shift_size=0,
mlp_ratio=4., qkv_bias=True, qk_scale=None, drop=0., attn_drop=0., drop_path=0.,
act_layer=GELU, norm_layer=nn.LayerNorm):
super().__init__()
self.dim = dim
self.input_resolution = input_resolution
self.num_heads = num_heads
self.window_size = window_size
self.shift_size = shift_size
self.mlp_ratio = mlp_ratio
if min(self.input_resolution) <= self.window_size:
self.shift_size = 0
self.window_size = min(self.input_resolution)
assert 0 <= self.shift_size < self.window_size, "shift_size must in 0-window_size"
self.norm1 = norm_layer(dim)
self.attn = WindowAttention(
dim,
window_size=[self.window_size, self.window_size],
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)
mlp_hidden_dim = int(dim * mlp_ratio)
self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop)
if self.shift_size > 0:
# ----------------------------------------------------------------#
# 由于进行特征提取时，会对输入的特征层进行的平移
# 如：
# [ [
# [1, 2, 3], [5, 6, 4],
# [4, 5, 6], --> [8, 9, 7],
# [7, 8, 9], [1, 2, 3],
# ] ]
# 这一步的作用就是使得平移后的区域块只计算自己部分的注意力机制
# ----------------------------------------------------------------#
H, W = self.input_resolution
_H, _W = _make_divisible(H, self.window_size), _make_divisible(W, self.window_size),
img_mask = torch.zeros((1, _H, _W, 1)) # 1 H W 1
h_slices = (slice(0, -self.window_size),
slice(-self.window_size, -self.shift_size),
slice(-self.shift_size, None))
w_slices = (slice(0, -self.window_size),
slice(-self.window_size, -self.shift_size),
slice(-self.shift_size, None))
cnt = 0
for h in h_slices:
for w in w_slices:
img_mask[:, h, w, :] = cnt
cnt += 1
mask_windows = window_partition(img_mask, self.window_size) # nW, window_size, window_size, 1
mask_windows = mask_windows.view(-1, self.window_size * self.window_size)
attn_mask = mask_windows.unsqueeze(1) - mask_windows.unsqueeze(2)
attn_mask = attn_mask.masked_fill(attn_mask != 0, float(-100.0)).masked_fill(attn_mask == 0, float(0.0))
self.attn_mask = attn_mask.cpu().numpy()
else:
self.attn_mask = None
def forward(self, x):
H, W = self.input_resolution
B, L, C = x.shape
assert L == H * W, "input feature has wrong size"
# -----------------------------------------------#
# bs, 3136, 96 -> bs, 56, 56, 96
# -----------------------------------------------#
shortcut = x
x = self.norm1(x)
x = x.view(B, H, W, C)
_H, _W = _make_divisible(H, self.window_size), _make_divisible(W, self.window_size),
x = x.permute(0, 3, 1, 2)
x = F.interpolate(x, [_H, _W], mode='bicubic', align_corners=False).permute(0, 2, 3, 1)
# -----------------------------------------------#
# 进行特征层的平移
# -----------------------------------------------#
if self.shift_size > 0:
shifted_x = torch.roll(x, shifts=(-self.shift_size, -self.shift_size), dims=(1, 2))
else:
shifted_x = x
# ------------------------------------------------------------------------------------------#
# bs, 56, 56, 96 -> bs * 64, 7, 7, 96 -> bs * 64, 49, 96
# ------------------------------------------------------------------------------------------#
x_windows = window_partition(shifted_x, self.window_size) # num_windows * B, window_size, window_size, C
x_windows = x_windows.view(-1, self.window_size * self.window_size, C) # nW*B, window_size*window_size, C
# -----------------------------------------------#
# bs * 64, 49, 97 -> bs * 64, 49, 97
# -----------------------------------------------#
if type(self.attn_mask) != type(None):
attn_mask = torch.tensor(self.attn_mask).cuda() if x.is_cuda else torch.tensor(self.attn_mask)
else:
attn_mask = None
attn_windows = self.attn(x_windows, mask=attn_mask) # nW*B, window_size*window_size, C
# -----------------------------------------------#
# bs * 64, 49, 97 -> bs, 56, 56, 96
# -----------------------------------------------#
attn_windows = attn_windows.view(-1, self.window_size, self.window_size, C)
shifted_x = window_reverse(attn_windows, self.window_size, _H, _W) # B H' W' C
# -----------------------------------------------#
# 将特征层平移回来
# -----------------------------------------------#
if self.shift_size > 0:
x = torch.roll(shifted_x, shifts=(self.shift_size, self.shift_size), dims=(1, 2))
else:
x = shifted_x
x = x.permute(0, 3, 1, 2)
x = F.interpolate(x, [H, W], mode='bicubic', align_corners=False).permute(0, 2, 3, 1)
# -----------------------------------------------#
# bs, 3136, 96
# -----------------------------------------------#
x = x.view(B, H * W, C)
# -----------------------------------------------#
# FFN
# bs, 3136, 96
# -----------------------------------------------#
x = shortcut + self.drop_path(x)
x = x + self.drop_path(self.mlp(self.norm2(x)))
return x
# 12.Swin-Transformer的基础模块:其实就是对 SwinTransformerBlock 模块取不同的参数，然后加一个下采样的选择
# -------------------------------------------------------#
# Swin-Transformer的基础模块。
# 使用窗口多头注意力机制进行特征提取。
# 使用PatchMerging进行高和宽的压缩。
# -------------------------------------------------------#
class BasicLayer(nn.Module):
def __init__(self, dim, input_resolution, depth, num_heads, window_size,
mlp_ratio=4., qkv_bias=True, qk_scale=None, drop=0., attn_drop=0.,
drop_path=0., norm_layer=nn.LayerNorm, downsample=None, use_checkpoint=False):
super().__init__()
# -------------------------------------------------------#
# 四个阶段对应不同的dim
# [96, 192, 384, 768]
# -------------------------------------------------------#
self.dim = dim
# -------------------------------------------------------#
# 四个阶段对应不同的输入分辨率
# [[56, 56], [28, 28], [14, 14], [7, 7]]
# -------------------------------------------------------#
self.input_resolution = input_resolution
# -------------------------------------------------------#
# 四个阶段对应不同的多头注意力机制重复次数
# [2, 2, 6, 2]
# -------------------------------------------------------#
self.depth = depth
self.use_checkpoint = use_checkpoint
# -------------------------------------------------------#
# 根据depth的次数利用窗口多头注意力机制进行特征提取。
# -------------------------------------------------------#
self.blocks = nn.ModuleList(
[
SwinTransformerBlock(
dim=dim,
input_resolution=input_resolution,
num_heads=num_heads,
window_size=window_size,
shift_size=0 if (i % 2 == 0) else window_size // 2,
mlp_ratio=mlp_ratio,
qkv_bias=qkv_bias,
qk_scale=qk_scale,
drop=drop,
attn_drop=attn_drop,
drop_path=drop_path[i] if isinstance(drop_path, list) else drop_path,
norm_layer=norm_layer
)
for i in range(depth)
]
)
if downsample is not None:
# -------------------------------------------------------#
# 判断是否要进行下采样，即：高宽压缩
# -------------------------------------------------------#
self.downsample = downsample(input_resolution, dim=dim, norm_layer=norm_layer)
else:
self.downsample = None
def forward(self, x):
for blk in self.blocks:
if self.use_checkpoint:
x_ = checkpoint.checkpoint(blk, x)
else:
x_ = blk(x)
if self.downsample is not None:
x = self.downsample(x_)
else:
x = x_
return x_, x
if __name__ == "__main__":
x = torch.randn(4, 3136, 96)
# net = SwinTransformerBlock(dim = 96, input_resolution = [56,56], num_heads = 8)
net = BasicLayer(dim=96, input_resolution=[56, 56],
depth=2, num_heads=3, window_size=7,
mlp_ratio=4., qkv_bias=True, qk_scale=None, drop=0., attn_drop=0.,
drop_path=0., norm_layer=nn.LayerNorm, downsample=None, use_checkpoint=False)
# net = BasicLayer(dim = [96, 192, 384, 768], input_resolution = [[56, 56], [28, 28], [14, 14], [7, 7]], depth=[2, 2, 6, 2], num_heads = [3, 6, 12, 24], window_size = 7,
# mlp_ratio=4., qkv_bias=True, qk_scale=None, drop=0., attn_drop=0.,
# drop_path=0., norm_layer=nn.LayerNorm, downsample=None, use_checkpoint=False)
outs = net(x)
print("10.====> ",len(outs), " ",outs[0].size())
运行结果如下：
作者：YANQ662
