130 lines
5.4 KiB
Python
130 lines
5.4 KiB
Python
#!/usr/bin/python3
|
|
#coding=utf-8
|
|
|
|
import torch
|
|
import numpy as np
|
|
import torch.nn.functional as F
|
|
from torch import nn
|
|
#import dataset
|
|
#from torch.utils import model_zoo
|
|
#from torchvision.models.resnet import resnet50
|
|
|
|
|
|
def weight_init_backbone(module):
|
|
for n, m in module.named_children():
|
|
#print('initialize: '+n)
|
|
if isinstance(m, nn.Conv2d):
|
|
nn.init.kaiming_normal_(m.weight, mode='fan_in', nonlinearity='relu')
|
|
if m.bias is not None:
|
|
nn.init.zeros_(m.bias)
|
|
elif isinstance(m, (nn.BatchNorm2d, nn.InstanceNorm2d)):
|
|
nn.init.ones_(m.weight)
|
|
if m.bias is not None:
|
|
nn.init.zeros_(m.bias)
|
|
elif isinstance(m, nn.Linear):
|
|
nn.init.kaiming_normal_(m.weight, mode='fan_in', nonlinearity='relu')
|
|
if m.bias is not None:
|
|
nn.init.zeros_(m.bias)
|
|
elif isinstance(m, nn.Sequential):
|
|
weight_init(m)
|
|
#elif isinstance(m, (nn.ReLU, nn.Sigmoid, ConvCaps, nn.PReLU, nn.AdaptiveAvgPool2d, nn.AdaptiveAvgPool1d, nn.Sigmoid, nn.Identity)):
|
|
elif isinstance(m, (nn.ReLU, nn.Sigmoid, nn.PReLU, nn.AdaptiveAvgPool2d, nn.AdaptiveAvgPool1d, nn.Sigmoid, nn.Identity)):
|
|
pass
|
|
else:
|
|
m.initialize()
|
|
|
|
def weight_init(module):
|
|
for n, m in module.named_children():
|
|
#print('initialize: '+n)
|
|
if isinstance(m, nn.Conv2d):
|
|
#nn.init.kaiming_normal_(m.weight, mode='fan_in', nonlinearity='relu')
|
|
nn.init.xavier_normal_(m.weight, gain=1)
|
|
if m.bias is not None:
|
|
nn.init.zeros_(m.bias)
|
|
elif isinstance(m, (nn.BatchNorm2d, nn.InstanceNorm2d)):
|
|
nn.init.ones_(m.weight)
|
|
if m.bias is not None:
|
|
nn.init.zeros_(m.bias)
|
|
elif isinstance(m, nn.Linear):
|
|
#nn.init.kaiming_normal_(m.weight, mode='fan_in', nonlinearity='relu')
|
|
nn.init.xavier_normal_(m.weight, gain=1)
|
|
if m.bias is not None:
|
|
nn.init.zeros_(m.bias)
|
|
elif isinstance(m, nn.Sequential):
|
|
weight_init(m)
|
|
#elif isinstance(m, (nn.ReLU, nn.Sigmoid, ConvCaps, nn.PReLU, nn.AdaptiveAvgPool2d, nn.AdaptiveAvgPool1d, nn.Sigmoid, nn.Identity)):
|
|
elif isinstance(m, (nn.ReLU, nn.Sigmoid, nn.PReLU, nn.AdaptiveAvgPool2d, nn.AdaptiveAvgPool1d, nn.Sigmoid, nn.Identity)):
|
|
pass
|
|
else:
|
|
m.initialize()
|
|
|
|
|
|
class Bottleneck(nn.Module):
|
|
def __init__(self, inplanes, planes, stride=1, downsample=None, dilation=1):
|
|
super(Bottleneck, self).__init__()
|
|
self.conv1 = nn.Conv2d(inplanes, planes, kernel_size=1, bias=False)
|
|
self.bn1 = nn.BatchNorm2d(planes)
|
|
self.conv2 = nn.Conv2d(planes, planes, kernel_size=3, stride=stride, padding=(3*dilation-1)//2, bias=False, dilation=dilation)
|
|
self.bn2 = nn.BatchNorm2d(planes)
|
|
self.conv3 = nn.Conv2d(planes, planes*4, kernel_size=1, bias=False)
|
|
self.bn3 = nn.BatchNorm2d(planes*4)
|
|
self.downsample = downsample
|
|
|
|
def forward(self, x):
|
|
out = F.relu(self.bn1(self.conv1(x)), inplace=True)
|
|
out = F.relu(self.bn2(self.conv2(out)), inplace=True)
|
|
out = self.bn3(self.conv3(out))
|
|
if self.downsample is not None:
|
|
x = self.downsample(x)
|
|
return F.relu(out+x, inplace=True)
|
|
|
|
class ResNet(nn.Module):
|
|
def __init__(self):
|
|
super(ResNet, self).__init__()
|
|
self.inplanes = 64
|
|
self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3, bias=False)
|
|
self.bn1 = nn.BatchNorm2d(64)
|
|
self.layer1 = self.make_layer( 64, 3, stride=1, dilation=1)
|
|
self.layer2 = self.make_layer(128, 4, stride=2, dilation=1)
|
|
self.layer3 = self.make_layer(256, 6, stride=2, dilation=1)
|
|
self.layer4 = self.make_layer(512, 3, stride=2, dilation=1)
|
|
#self.initialize()
|
|
|
|
def make_layer(self, planes, blocks, stride, dilation):
|
|
downsample = nn.Sequential(nn.Conv2d(self.inplanes, planes*4, kernel_size=1, stride=stride, bias=False), nn.BatchNorm2d(planes*4))
|
|
layers = [Bottleneck(self.inplanes, planes, stride, downsample, dilation=dilation)]
|
|
self.inplanes = planes*4
|
|
for _ in range(1, blocks):
|
|
layers.append(Bottleneck(self.inplanes, planes, dilation=dilation))
|
|
return nn.Sequential(*layers)
|
|
|
|
def forward(self, x):
|
|
out1 = F.relu(self.bn1(self.conv1(x)), inplace=True)
|
|
out1 = F.max_pool2d(out1, kernel_size=3, stride=2, padding=1)
|
|
out2 = self.layer1(out1)
|
|
out3 = self.layer2(out2)
|
|
out4 = self.layer3(out3)
|
|
out5 = self.layer4(out4)
|
|
# 1/4,1/8,1/16,1/32
|
|
return out2,out3,out4,out5
|
|
|
|
def initialize(self):
|
|
self.load_state_dict(torch.load('Weights/resnet50.pth'), strict=False)
|
|
|
|
'''
|
|
|
|
if __name__ == "__main__":
|
|
cfg = dataset.Config(datapath='datasets/DUTS', savepath='checkpoint/ICON/ICON-S', mode='train', batch=32, lr=0.05, momen=0.9, decay=5e-4, epoch=48)
|
|
model = ResNet(cfg)
|
|
#pretrained_dict = model_zoo.load_url(model_urls['resnet50'])
|
|
#model_dict = model.state_dict()
|
|
#pretrained_dict = {k: v for k, v in pretrained_dict.items() if k in model_dict}
|
|
#model_dict.update(pretrained_dict)
|
|
#model.load_state_dict(model_dict)
|
|
input = torch.autograd.Variable(torch.randn(4, 3, 384, 384))
|
|
output = model(input)
|
|
print(output[1].size())
|
|
'''
|
|
|
|
|