Pytorch-HED fine-tune实现
Pytorch HED （VGG16-HED and Res34-HED ）
python 3 ； pytorch 0.4
基于 fine-tune 的 VGG16 或者 Resnet34 构建HED网络；并且，本人基于pytorch已经写好和训练好的网络基础上，构建属于自己简约风格的网络，例如HED网络。
https://github.com/gengyanlei 链接 具体情况
''' VGG16 network
Sequential(
(0): Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): ReLU(inplace)
(2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(3): ReLU(inplace)
(4): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(5): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(6): ReLU(inplace)
(7): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(8): ReLU(inplace)
(9): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(10): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(11): ReLU(inplace)
(12): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(13): ReLU(inplace)
(14): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(15): ReLU(inplace)
(16): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(17): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(18): ReLU(inplace)
(19): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(20): ReLU(inplace)
(21): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(22): ReLU(inplace)
(23): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(24): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(25): ReLU(inplace)
(26): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(27): ReLU(inplace)
(28): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(29): ReLU(inplace)
(30): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
)
'''
import torch
import torchvision
from torch import nn
# input size [256,256]
# 基于vgg16 hed
class HED_vgg16(nn.Module):
def __init__(self,num_filters=32, pretrained=False,class_number=2):
# Here is the function part, with no braces ()
super().__init__()
self.encoder = torchvision.models.vgg16(pretrained=pretrained).features
self.pool = nn.MaxPool2d(2, 2)
self.conv1=self.encoder[0:4]
self.score1=nn.Sequential(nn.Conv2d(num_filters*2,1,1,1),nn.ReLU())# 256*256
self.conv2=self.encoder[5:9]
self.d_conv2=nn.Sequential(nn.Conv2d(num_filters*4,1,1,1),nn.ReLU())#128*128
self.score2=nn.UpsamplingBilinear2d(scale_factor=2)#256*256
self.conv3=self.encoder[10:16]
self.d_conv3=nn.Sequential(nn.Conv2d(num_filters*8,1,1,1),nn.ReLU())#64*64
self.score3=nn.UpsamplingBilinear2d(scale_factor=4)#256*256
self.conv4=self.encoder[17:23]
self.d_conv4=nn.Sequential(nn.Conv2d(num_filters*16,1,1,1),nn.ReLU())#32*32
self.score4=nn.UpsamplingBilinear2d(scale_factor=8)#256*256
self.conv5=self.encoder[24:30]
self.d_conv5=nn.Sequential(nn.Conv2d(num_filters*16,1,1,1),nn.ReLU())#16*16
self.score5=nn.UpsamplingBilinear2d(scale_factor=16)#256*256
self.score=nn.Conv2d(5,class_number,1,1)# No relu
def forward(self,x):
# Here is the part that calculates the return value
x=self.conv1(x)
s1=self.score1(x)
x=self.pool(x)
x=self.conv2(x)
s_x=self.d_conv2(x)
s2=self.score2(s_x)
x=self.pool(x)
x=self.conv3(x)
s_x=self.d_conv3(x)
s3=self.score3(s_x)
x=self.pool(x)
x=self.conv3(x)
s_x=self.d_conv4(x)
s4=self.score4(s_x)
x=self.pool(x)
x=self.conv5(x)
s_x=self.d_conv5(x)
s5=self.score5(s_x)
score=self.score(torch.cat([s1,s2,s3,s4,s5],axis=1))
return score
''' you need to write softmax after model and predict output by yourself '''
hed1=HED_vgg16()
print(hed1)
print(hed1.state_dict().keys())
######################################################################
#import torch
#import torchvision
#from torch import nn
# 基于resnet34 hed
class HED_res34(nn.Module):
def __init__(self,num_filters=32, pretrained=False,class_number=2):
super().__init__()
self.encoder = torchvision.models.resnet34(pretrained=pretrained)
self.pool = nn.MaxPool2d(3,2,1)
#start
self.start=nn.Sequential(self.encoder.conv1,self.encoder.bn1,self.encoder.relu)#128*128
self.d_convs=nn.Sequential(nn.Conv2d(num_filters*2,1,1,1),nn.ReLU())
self.scores=nn.UpsamplingBilinear2d(scale_factor=2)#256*256
self.layer1=self.encoder.layer1#64*64
self.d_conv1=nn.Sequential(nn.Conv2d(num_filters*2,1,1,1),nn.ReLU())
self.score1=nn.UpsamplingBilinear2d(scale_factor=4)#256*256
self.layer2=self.encoder.layer2#32*32
self.d_conv2=nn.Sequential(nn.Conv2d(num_filters*4,1,1,1),nn.ReLU())
self.score2=nn.UpsamplingBilinear2d(scale_factor=8)#256*256
self.layer3=self.encoder.layer3#16*16
self.d_conv3=nn.Sequential(nn.Conv2d(num_filters*8,1,1,1),nn.ReLU())
self.score3=nn.UpsamplingBilinear2d(scale_factor=16)#256*256
self.layer4=self.encoder.layer4#8*8
self.d_conv4=nn.Sequential(nn.Conv2d(num_filters*16,1,1,1),nn.ReLU())
self.score4=nn.UpsamplingBilinear2d(scale_factor=32)#256*256
self.score=nn.Conv2d(5,class_number,1,1)# No relu loss_func has softmax
def forward(self,x):
x=self.start(x)
s_x=self.d_convs(x)
ss=self.scores(s_x)
x=self.pool(x)
x=self.layer1(x)
s_x=self.d_conv1(x)
s1=self.score1(s_x)
x=self.layer2(x)
s_x=self.d_conv2(x)
s2=self.score2(s_x)
x=self.layer3(x)
s_x=self.d_conv3(x)
s3=self.score3(s_x)
x=self.layer4(x)
s_x=self.d_conv4(x)
s4=self.score4(s_x)
score=self.score(torch.cat([s1,s2,s3,s4,ss],axis=1))
return score
hed2=HED_res34()
print(hed2)
print(hed2.state_dict().keys())