百度框架下的最轻量化模型 PP-LCNet，Pytorch版本代码
近年来，有很多轻量级的骨干网络问世，尤其最近两年，各种 NAS 搜索出的网络层出不穷，这些网络要么主打 FLOPs 或者 Params 上的优势，要么主打 ARM 设备上的推理速度的优势，很少有网络专门针对 Intel CPU 做特定的优化，导致这些网络在 Intel CPU 端的推理速度并不是很完美。基于此，我们针对 Intel CPU 设备以及其加速库 MKLDNN 设计了特定的骨干网络 PP-LCNet，比起其他的轻量级的 SOTA 模型，该骨干网络可以在不增加推理时间的情况下，进一步提升模型的性能，最终大幅度超越现有的 SOTA 模型。与其他模型的对比图如下。
import torch
import torch.nn as nn
NET_CONFIG = {
"blocks2":
#k, in_c, out_c, s, use_se
[[3, 16, 32, 1, False]],
"blocks3": [[3, 32, 64, 2, False], [3, 64, 64, 1, False]],
"blocks4": [[3, 64, 128, 2, False], [3, 128, 128, 1, False]],
"blocks5": [[3, 128, 256, 2, False], [5, 256, 256, 1, False],
[5, 256, 256, 1, False], [5, 256, 256, 1, False],
[5, 256, 256, 1, False], [5, 256, 256, 1, False]],
"blocks6": [[5, 256, 512, 2, True], [5, 512, 512, 1, True]]
}
def make_divisible(v, divisor=8, 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
class ConvBNLayer(nn.Module):
def __init__(self,
num_channels,
filter_size,
num_filters,
stride,
num_groups=1):
super().__init__()
self.conv = nn.Conv2d(
in_channels=num_channels,
out_channels=num_filters,
kernel_size=filter_size,
stride=stride,
padding=(filter_size - 1) // 2,
groups=num_groups,
)
self.bn = nn.BatchNorm2d(num_filters)
self.hardswish = nn.Hardswish()
def forward(self, x):
x = self.conv(x)
x = self.bn(x)
x = self.hardswish(x)
return x
class SEModule(nn.Module):
def __init__(self, channel, reduction=4):
super().__init__()
self.avg_pool = nn.AdaptiveAvgPool2d(1)
self.conv1 = nn.Conv2d(
in_channels=channel,
out_channels=channel // reduction,
kernel_size=1,
stride=1,
padding=0)
self.relu = nn.ReLU()
self.conv2 = nn.Conv2d(
in_channels=channel // reduction,
out_channels=channel,
kernel_size=1,
stride=1,
padding=0)
self.hardsigmoid = nn.Hardsigmoid()
def forward(self, x):
identity = x
x = self.avg_pool(x)
x = self.conv1(x)
x = self.relu(x)
x = self.conv2(x)
x = self.hardsigmoid(x)
x = identity*x
return x
class DepthwiseSeparable(nn.Module):
def __init__(self,
num_channels,
num_filters,
stride,
dw_size=3,
use_se=False):
super().__init__()
self.use_se = use_se
self.dw_conv = ConvBNLayer(
num_channels=num_channels,
num_filters=num_channels,
filter_size=dw_size,
stride=stride,
num_groups=num_channels)
if use_se:
self.se = SEModule(num_channels)
self.pw_conv = ConvBNLayer(
num_channels=num_channels,
filter_size=1,
num_filters=num_filters,
stride=1)
def forward(self, x):
x = self.dw_conv(x)
if self.use_se:
x = self.se(x)
x = self.pw_conv(x)
return x
class PP_LCNet(nn.Module):
def __init__(self,
scale=1.0,
class_num=1000,
dropout_prob=0.2,
class_expand=1280,
return_patterns=None,
return_stages=None):
super().__init__()
self.scale = scale
self.class_expand = class_expand
self.conv1 = ConvBNLayer(
num_channels=3,
filter_size=3,
num_filters=make_divisible(16 * scale),
stride=2)
self.blocks2 = nn.Sequential(* [
DepthwiseSeparable(
num_channels=make_divisible(in_c * scale),
num_filters=make_divisible(out_c * scale),
dw_size=k,
stride=s,
use_se=se)
for i, (k, in_c, out_c, s, se) in enumerate(NET_CONFIG["blocks2"])
])
self.blocks3 = nn.Sequential(* [
DepthwiseSeparable(
num_channels=make_divisible(in_c * scale),
num_filters=make_divisible(out_c * scale),
dw_size=k,
stride=s,
use_se=se)
for i, (k, in_c, out_c, s, se) in enumerate(NET_CONFIG["blocks3"])
])
self.blocks4 = nn.Sequential(* [
DepthwiseSeparable(
num_channels=make_divisible(in_c * scale),
num_filters=make_divisible(out_c * scale),
dw_size=k,
stride=s,
use_se=se)
for i, (k, in_c, out_c, s, se) in enumerate(NET_CONFIG["blocks4"])
])
self.blocks5 = nn.Sequential(* [
DepthwiseSeparable(
num_channels=make_divisible(in_c * scale),
num_filters=make_divisible(out_c * scale),
dw_size=k,
stride=s,
use_se=se)
for i, (k, in_c, out_c, s, se) in enumerate(NET_CONFIG["blocks5"])
])
self.blocks6 = nn.Sequential(* [
DepthwiseSeparable(
num_channels=make_divisible(in_c * scale),
num_filters=make_divisible(out_c * scale),
dw_size=k,
stride=s,
use_se=se)
for i, (k, in_c, out_c, s, se) in enumerate(NET_CONFIG["blocks6"])
])
self.avg_pool = nn.AdaptiveAvgPool2d(1)
self.last_conv = nn.Conv2d(
in_channels=make_divisible(NET_CONFIG["blocks6"][-1][2] * scale),
out_channels=self.class_expand,
kernel_size=1,
stride=1,
padding=0,)
self.hardswish = nn.Hardswish()
self.dropout = nn.Dropout(p=dropout_prob)
self.fc = nn.Linear(self.class_expand, class_num)
def forward(self, x):
x = self.conv1(x)
x = self.blocks2(x)
x = self.blocks3(x)
x = self.blocks4(x)
x = self.blocks5(x)
x = self.blocks6(x)
x = self.avg_pool(x)
x = self.last_conv(x)
x = self.hardswish(x)
x = self.dropout(x)
x = torch.flatten(x, start_dim=1, end_dim=-1)
x = self.fc(x)
return x
def PPLCNET_x0_25(**kwargs):
model = PP_LCNet(scale=0.25, **kwargs)
return model
def PPLCNET_x0_35(**kwargs):
model = PP_LCNet(scale=0.35, **kwargs)
return model
def PPLCNET_x0_5(**kwargs):
model = PP_LCNet(scale=0.5, **kwargs)
return model
def PPLCNET_x0_75(**kwargs):
model = PP_LCNet(scale=0.75, **kwargs)
return model
def PPLCNET_x1_0(**kwargs):
model = PP_LCNet(scale=1.0, **kwargs)
return model
def PPLCNET_x1_5(**kwargs):
model = PP_LCNet(scale=1.5, **kwargs)
return model
def PPLCNET_x2_0(**kwargs):
model = PP_LCNet(scale=2.0, **kwargs)
return model
def PPLCNET_x2_5(**kwargs):
model = PP_LCNet(scale=2.5, **kwargs)
return model
if __name__ == '__main__':
model = PPLCNET_x1_5()
input = torch.randn(1, 3, 224, 224)
print(input.shape)
output = model(input)
print(output.shape)