DSAN代码:Deep Subdomain Adaptation Network forImage Classification
DSAN.py
import torch
import torch.nn as nn
import ResNet
import lmmd
class DSAN(nn.Module):
def __init__(self, num_classes=31, bottle_neck=True):
super(DSAN, self).__init__()
self.feature_layers = ResNet.resnet50(True)
self.lmmd_loss = lmmd.LMMD_loss(class_num=num_classes)
self.bottle_neck = bottle_neck
if bottle_neck:
self.bottle = nn.Linear(2048, 256)
self.cls_fc = nn.Linear(256, num_classes)
else:
self.cls_fc = nn.Linear(2048, num_classes)
def forward(self, source, target, s_label):
source = self.feature_layers(source)
if self.bottle_neck:
source = self.bottle(source)
s_pred = self.cls_fc(source)
target = self.feature_layers(target)
if self.bottle_neck:
target = self.bottle(target)
t_label = self.cls_fc(target)
loss_lmmd = self.lmmd_loss.get_loss(source, target, s_label, torch.nn.functional.softmax(t_label, dim=1))
return s_pred, loss_lmmd
def predict(self, x):
x = self.feature_layers(x)
if self.bottle_neck:
x = self.bottle(x)
return self.cls_fc(x)ResNet.py
import torch.nn as nn
import math
import torch.utils.model_zoo as model_zoo
__all__ = ['ResNet', 'resnet18', 'resnet34', 'resnet50', 'resnet101',
'resnet152']
model_urls = {
'resnet18': 'https://download.pytorch.org/models/resnet18-5c106cde.pth',
'resnet34': 'https://download.pytorch.org/models/resnet34-333f7ec4.pth',
'resnet50': 'https://download.pytorch.org/models/resnet50-19c8e357.pth',
'resnet101': 'https://download.pytorch.org/models/resnet101-5d3b4d8f.pth',
'resnet152': 'https://download.pytorch.org/models/resnet152-b121ed2d.pth',
}
def conv3x3(in_planes, out_planes, stride=1):
"""3x3 convolution with padding"""
return nn.Conv2d(in_planes, out_planes, kernel_size=3, stride=stride,
padding=1, bias=False)
class BasicBlock(nn.Module):
expansion = 1
def __init__(self, inplanes, planes, stride=1, downsample=None):
super(BasicBlock, self).__init__()
self.conv1 = conv3x3(inplanes, planes, stride)
self.bn1 = nn.BatchNorm2d(planes)
self.relu = nn.ReLU(inplace=True)
self.conv2 = conv3x3(planes, planes)
self.bn2 = nn.BatchNorm2d(planes)
self.downsample = downsample
self.stride = stride
def forward(self, x):
residual = x
out = self.conv1(x)
out = self.bn1(out)
out = self.relu(out)
out = self.conv2(out)
out = self.bn2(out)
if self.downsample is not None:
residual = self.downsample(x)
out += residual
out = self.relu(out)
return out
class Bottleneck(nn.Module):
expansion = 4
def __init__(self, inplanes, planes, stride=1, downsample=None):
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=1, bias=False)
self.bn2 = nn.BatchNorm2d(planes)
self.conv3 = nn.Conv2d(planes, planes * 4, kernel_size=1, bias=False)
self.bn3 = nn.BatchNorm2d(planes * 4)
self.relu = nn.ReLU(inplace=True)
self.downsample = downsample
self.stride = stride
def forward(self, x):
residual = x
out = self.conv1(x)
out = self.bn1(out)
out = self.relu(out)
out = self.conv2(out)
out = self.bn2(out)
out = self.relu(out)
out = self.conv3(out)
out = self.bn3(out)
if self.downsample is not None:
residual = self.downsample(x)
out += residual
out = self.relu(out)
return out
class ResNet(nn.Module):
def __init__(self, block, layers, num_classes=1000):
self.inplanes = 64
super(ResNet, self).__init__()
self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3,
bias=False)
self.bn1 = nn.BatchNorm2d(64)
self.relu = nn.ReLU(inplace=True)
self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
self.layer1 = self._make_layer(block, 64, layers[0])
self.layer2 = self._make_layer(block, 128, layers[1], stride=2)
self.layer3 = self._make_layer(block, 256, layers[2], stride=2)
self.layer4 = self._make_layer(block, 512, layers[3], stride=2)
self.avgpool = nn.AvgPool2d(7, stride=1)
self.baselayer = [self.conv1, self.bn1, self.layer1, self.layer2, self.layer3, self.layer4]
self.fc = nn.Linear(512 * block.expansion, num_classes)
for m in self.modules():
if isinstance(m, nn.Conv2d):
n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
m.weight.data.normal_(0, math.sqrt(2. / n))
elif isinstance(m, nn.BatchNorm2d):
m.weight.data.fill_(1)
m.bias.data.zero_()
def _make_layer(self, block, planes, blocks, stride=1):
downsample = None
if stride != 1 or self.inplanes != planes * block.expansion:
downsample = nn.Sequential(
nn.Conv2d(self.inplanes, planes * block.expansion,
kernel_size=1, stride=stride, bias=False),
nn.BatchNorm2d(planes * block.expansion),
)
layers = []
layers.append(block(self.inplanes, planes, stride, downsample))
self.inplanes = planes * block.expansion
for i in range(1, blocks):
layers.append(block(self.inplanes, planes))
return nn.Sequential(*layers)
def forward(self, x):
x = self.conv1(x)
x = self.bn1(x)
x = self.relu(x)
x = self.maxpool(x)
x = self.layer1(x)
x = self.layer2(x)
x = self.layer3(x)
x = self.layer4(x)
x = self.avgpool(x)
x = x.view(x.size(0), -1)
return x
def resnet50(pretrained=False, **kwargs):
"""Constructs a ResNet-50 model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
model = ResNet(Bottleneck, [3, 4, 6, 3], **kwargs)
if pretrained:
model.load_state_dict(model_zoo.load_url(model_urls['resnet50']))
return modeldata_loader.py
from torchvision import datasets, transforms
import torch
import os
def load_training(root_path, dir, batch_size, kwargs):
transform = transforms.Compose(
[transforms.Resize([256, 256]),
transforms.RandomCrop(224),
transforms.RandomHorizontalFlip(),
transforms.ToTensor()])
data = datasets.ImageFolder(root=os.path.join(root_path, dir), transform=transform)
train_loader = torch.utils.data.DataLoader(data, batch_size=batch_size, shuffle=True, drop_last=True, **kwargs)
return train_loader
def load_testing(root_path, dir, batch_size, kwargs):
transform = transforms.Compose(
[transforms.Resize([224, 224]),
transforms.ToTensor()])
data = datasets.ImageFolder(root=os.path.join(root_path, dir), transform=transform)
test_loader = torch.utils.data.DataLoader(data, batch_size=batch_size, shuffle=True, **kwargs)
return test_loaderImmd.py
import torch
import torch.nn as nn
import numpy as np
class LMMD_loss(nn.Module):
def __init__(self, class_num=31, kernel_type='rbf', kernel_mul=2.0, kernel_num=5, fix_sigma=None):
super(LMMD_loss, self).__init__()
self.class_num = class_num
self.kernel_num = kernel_num
self.kernel_mul = kernel_mul
self.fix_sigma = fix_sigma
self.kernel_type = kernel_type
def guassian_kernel(self, source, target, kernel_mul=2.0, kernel_num=5, fix_sigma=None):
n_samples = int(source.size()[0]) + int(target.size()[0])
total = torch.cat([source, target], dim=0)
total0 = total.unsqueeze(0).expand(
int(total.size(0)), int(total.size(0)), int(total.size(1)))
total1 = total.unsqueeze(1).expand(
int(total.size(0)), int(total.size(0)), int(total.size(1)))
L2_distance = ((total0-total1)**2).sum(2)
if fix_sigma:
bandwidth = fix_sigma
else:
bandwidth = torch.sum(L2_distance.data) / (n_samples**2-n_samples)
bandwidth /= kernel_mul ** (kernel_num // 2)
bandwidth_list = [bandwidth * (kernel_mul**i)
for i in range(kernel_num)]
kernel_val = [torch.exp(-L2_distance / bandwidth_temp)
for bandwidth_temp in bandwidth_list]
return sum(kernel_val)
def get_loss(self, source, target, s_label, t_label):
batch_size = source.size()[0]
weight_ss, weight_tt, weight_st = self.cal_weight(
s_label, t_label, batch_size=batch_size, class_num=self.class_num)
weight_ss = torch.from_numpy(weight_ss).cuda()
weight_tt = torch.from_numpy(weight_tt).cuda()
weight_st = torch.from_numpy(weight_st).cuda()
kernels = self.guassian_kernel(source, target,
kernel_mul=self.kernel_mul, kernel_num=self.kernel_num, fix_sigma=self.fix_sigma)
loss = torch.Tensor([0]).cuda()
if torch.sum(torch.isnan(sum(kernels))):
return loss
SS = kernels[:batch_size, :batch_size]
TT = kernels[batch_size:, batch_size:]
ST = kernels[:batch_size, batch_size:]
loss += torch.sum(weight_ss * SS + weight_tt * TT - 2 * weight_st * ST)
return loss
def convert_to_onehot(self, sca_label, class_num=31):
return np.eye(class_num)[sca_label]
def cal_weight(self, s_label, t_label, batch_size=32, class_num=31):
batch_size = s_label.size()[0]
s_sca_label = s_label.cpu().data.numpy()
s_vec_label = self.convert_to_onehot(s_sca_label, class_num=self.class_num)
s_sum = np.sum(s_vec_label, axis=0).reshape(1, class_num)
s_sum[s_sum == 0] = 100
s_vec_label = s_vec_label / s_sum
t_sca_label = t_label.cpu().data.max(1)[1].numpy()
t_vec_label = t_label.cpu().data.numpy()
t_sum = np.sum(t_vec_label, axis=0).reshape(1, class_num)
t_sum[t_sum == 0] = 100
t_vec_label = t_vec_label / t_sum
index = list(set(s_sca_label) & set(t_sca_label))
mask_arr = np.zeros((batch_size, class_num))
mask_arr[:, index] = 1
t_vec_label = t_vec_label * mask_arr
s_vec_label = s_vec_label * mask_arr
weight_ss = np.matmul(s_vec_label, s_vec_label.T)
weight_tt = np.matmul(t_vec_label, t_vec_label.T)
weight_st = np.matmul(s_vec_label, t_vec_label.T)
length = len(index)
if length != 0:
weight_ss = weight_ss / length
weight_tt = weight_tt / length
weight_st = weight_st / length
else:
weight_ss = np.array([0])
weight_tt = np.array([0])
weight_st = np.array([0])
return weight_ss.astype('float32'), weight_tt.astype('float32'), weight_st.astype('float32')main.py
import torch import torch.nn.functional as F import math import argparse import numpy as np import os from DSAN import DSAN import data_loader def load_data(root_path, src, tar, batch_size): kwargs = {'num_workers': 1, 'pin_memory': True} loader_src = data_loader.load_training(root_path, src, batch_size, kwargs) loader_tar = data_loader.load_training(root_path, tar, batch_size, kwargs) loader_tar_test = data_loader.load_testing( root_path, tar, batch_size, kwargs) return loader_src, loader_tar, loader_tar_test def train_epoch(epoch, model, dataloaders, optimizer): model.train() source_loader, target_train_loader, _ = dataloaders iter_source = iter(source_loader) iter_target = iter(target_train_loader) num_iter = len(source_loader) for i in range(1, num_iter): data_source, label_source = iter_source.next() data_target, _ = iter_target.next() if i % len(target_train_loader) == 0: iter_target = iter(target_train_loader) data_source, label_source = data_source.cuda(), label_source.cuda() data_target = data_target.cuda() optimizer.zero_grad() label_source_pred, loss_lmmd = model( data_source, data_target, label_source) loss_cls = F.nll_loss(F.log_softmax( label_source_pred, dim=1), label_source) lambd = 2 / (1 + math.exp(-10 * (epoch) / args.nepoch)) - 1 loss = loss_cls + args.weight * lambd * loss_lmmd loss.backward() optimizer.step() if i % args.log_interval == 0: print(f'Epoch: [{epoch:2d}], Loss: {loss.item():.4f}, cls_Loss: {loss_cls.item():.4f}, loss_lmmd: {loss_lmmd.item():.4f}') def test(model, dataloader): model.eval() test_loss = 0 correct = 0 with torch.no_grad(): for data, target in dataloader: data, target = data.cuda(), target.cuda() pred = model.predict(data) # sum up batch loss test_loss += F.nll_loss(F.log_softmax(pred, dim=1), target).item() pred = pred.data.max(1)[1] correct += pred.eq(target.data.view_as(pred)).cpu().sum() test_loss /= len(dataloader) print( f'Average loss: {test_loss:.4f}, Accuracy: {correct}/{len(dataloader.dataset)} ({100. * correct / len(dataloader.dataset):.2f}%)') return correct def get_args(): def str2bool(v): if v.lower() in ('yes', 'true', 't', 'y', '1'): return True elif v.lower() in ('no', 'false', 'f', 'n', '0'): return False else: raise argparse.ArgumentTypeError('Unsupported value encountered.') parser = argparse.ArgumentParser() parser.add_argument('--root_path', type=str, help='Root path for dataset', default='/data/zhuyc/OFFICE31/') parser.add_argument('--src', type=str, help='Source domain', default='amazon') parser.add_argument('--tar', type=str, help='Target domain', default='webcam') parser.add_argument('--nclass', type=int, help='Number of classes', default=31) parser.add_argument('--batch_size', type=float, help='batch size', default=32) parser.add_argument('--nepoch', type=int, help='Total epoch num', default=200) parser.add_argument('--lr', type=list, help='Learning rate', default=[0.001, 0.01, 0.01]) parser.add_argument('--early_stop', type=int, help='Early stoping number', default=30) parser.add_argument('--seed', type=int, help='Seed', default=2021) parser.add_argument('--weight', type=float, help='Weight for adaptation loss', default=0.5) parser.add_argument('--momentum', type=float, help='Momentum', default=0.9) parser.add_argument('--decay', type=float, help='L2 weight decay', default=5e-4) parser.add_argument('--bottleneck', type=str2bool, nargs='?', const=True, default=True) parser.add_argument('--log_interval', type=int, help='Log interval', default=10) parser.add_argument('--gpu', type=str, help='GPU ID', default='0') args = parser.parse_args() return args if __name__ == '__main__': args = get_args() print(vars(args)) SEED = args.seed np.random.seed(SEED) torch.manual_seed(SEED) torch.cuda.manual_seed_all(SEED) torch.backends.cudnn.deterministic = True torch.backends.cudnn.benchmark = False os.environ["CUDA_VISIBLE_DEVICES"] = args.gpu dataloaders = load_data(args.root_path, args.src, args.tar, args.batch_size) model = DSAN(num_classes=args.nclass).cuda() correct = 0 stop = 0 if args.bottleneck: optimizer = torch.optim.SGD([ {'params': model.feature_layers.parameters()}, {'params': model.bottle.parameters(), 'lr': args.lr[1]}, {'params': model.cls_fc.parameters(), 'lr': args.lr[2]}, ], lr=args.lr[0], momentum=args.momentum, weight_decay=args.decay) else: optimizer = torch.optim.SGD([ {'params': model.feature_layers.parameters()}, {'params': model.cls_fc.parameters(), 'lr': args.lr[1]}, ], lr=args.lr[0], momentum=args.momentum, weight_decay=args.decay) for epoch in range(1, args.nepoch + 1): stop += 1 for index, param_group in enumerate(optimizer.param_groups): param_group['lr'] = args.lr[index] / math.pow((1 + 10 * (epoch - 1) / args.nepoch), 0.75) train_epoch(epoch, model, dataloaders, optimizer) t_correct = test(model, dataloaders[-1]) if t_correct > correct: correct = t_correct stop = 0 torch.save(model, 'model.pkl') print( f'{args.src}-{args.tar}: max correct: {correct} max accuracy: {100. * correct / len(dataloaders[-1].dataset):.2f}%\n') if stop >= args.early_stop: print( f'Final test acc: {100. * correct / len(dataloaders[-1].dataset):.2f}%') break