diff --git a/.gitignore b/.gitignore
new file mode 100644
index 0000000..a9545d8
--- /dev/null
+++ b/.gitignore
@@ -0,0 +1,3 @@
+*.pyc
+*.pth
+/__pycache__
diff --git a/README.md b/README.md
index 8619899..a3660ef 100644
--- a/README.md
+++ b/README.md
@@ -1,3 +1,2 @@
-# -BUAA_FRB_-
-
-这里是由阳神带领的冯如小队，完毕！
\ No newline at end of file
+# FRB_lttprivate
+save code of myself
diff --git a/deeplab.py b/deeplab.py
index 50b1e90..2c58938 100644
--- a/deeplab.py
+++ b/deeplab.py
@@ -61,6 +61,14 @@ def get_10x_lr_params(self):
                             yield p
 
 
+def getinput():
+   root_path = "../data20200111/"
+   img_path = "img/"
+   fea_path = "feature/"
+   label_path = "label/"
+
+
+
 if __name__ == "__main__":
     model = DeepLab(backbone='mobilenet', output_stride=16)
     model.eval()
diff --git a/modeling/__init__.py b/modeling/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/modeling/aspp.py b/modeling/aspp.py
new file mode 100644
index 0000000..5a97879
--- /dev/null
+++ b/modeling/aspp.py
@@ -0,0 +1,95 @@
+import math
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from modeling.sync_batchnorm.batchnorm import SynchronizedBatchNorm2d
+
+class _ASPPModule(nn.Module):
+    def __init__(self, inplanes, planes, kernel_size, padding, dilation, BatchNorm):
+        super(_ASPPModule, self).__init__()
+        self.atrous_conv = nn.Conv2d(inplanes, planes, kernel_size=kernel_size,
+                                            stride=1, padding=padding, dilation=dilation, bias=False)
+        self.bn = BatchNorm(planes)
+        self.relu = nn.ReLU()
+
+        self._init_weight()
+
+    def forward(self, x):
+        x = self.atrous_conv(x)
+        x = self.bn(x)
+
+        return self.relu(x)
+
+    def _init_weight(self):
+        for m in self.modules():
+            if isinstance(m, nn.Conv2d):
+                torch.nn.init.kaiming_normal_(m.weight)
+            elif isinstance(m, SynchronizedBatchNorm2d):
+                m.weight.data.fill_(1)
+                m.bias.data.zero_()
+            elif isinstance(m, nn.BatchNorm2d):
+                m.weight.data.fill_(1)
+                m.bias.data.zero_()
+
+class ASPP(nn.Module):
+    def __init__(self, backbone, output_stride, BatchNorm):
+        super(ASPP, self).__init__()
+        if backbone == 'drn':
+            inplanes = 512
+        elif backbone == 'mobilenet':
+            inplanes = 320
+        else:
+            inplanes = 2048
+        if output_stride == 16:
+            dilations = [1, 6, 12, 18]
+        elif output_stride == 8:
+            dilations = [1, 12, 24, 36]
+        else:
+            raise NotImplementedError
+
+        self.aspp1 = _ASPPModule(inplanes, 256, 1, padding=0, dilation=dilations[0], BatchNorm=BatchNorm)
+        self.aspp2 = _ASPPModule(inplanes, 256, 3, padding=dilations[1], dilation=dilations[1], BatchNorm=BatchNorm)
+        self.aspp3 = _ASPPModule(inplanes, 256, 3, padding=dilations[2], dilation=dilations[2], BatchNorm=BatchNorm)
+        self.aspp4 = _ASPPModule(inplanes, 256, 3, padding=dilations[3], dilation=dilations[3], BatchNorm=BatchNorm)
+
+        self.global_avg_pool = nn.Sequential(nn.AdaptiveAvgPool2d((1, 1)),
+                                             nn.Conv2d(inplanes, 256, 1, stride=1, bias=False),
+                                             BatchNorm(256),
+                                             nn.ReLU())
+        self.conv1 = nn.Conv2d(1280, 256, 1, bias=False)
+        self.bn1 = BatchNorm(256)
+        self.relu = nn.ReLU()
+        self.dropout = nn.Dropout(0.5)
+        self._init_weight()
+
+    def forward(self, x):
+        x1 = self.aspp1(x)
+        x2 = self.aspp2(x)
+        x3 = self.aspp3(x)
+        x4 = self.aspp4(x)
+        x5 = self.global_avg_pool(x)
+        x5 = F.interpolate(x5, size=x4.size()[2:], mode='bilinear', align_corners=True)
+        x = torch.cat((x1, x2, x3, x4, x5), dim=1)
+
+        x = self.conv1(x)
+        x = self.bn1(x)
+        x = self.relu(x)
+
+        return self.dropout(x)
+
+    def _init_weight(self):
+        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))
+                torch.nn.init.kaiming_normal_(m.weight)
+            elif isinstance(m, SynchronizedBatchNorm2d):
+                m.weight.data.fill_(1)
+                m.bias.data.zero_()
+            elif isinstance(m, nn.BatchNorm2d):
+                m.weight.data.fill_(1)
+                m.bias.data.zero_()
+
+
+def build_aspp(backbone, output_stride, BatchNorm):
+    return ASPP(backbone, output_stride, BatchNorm)
\ No newline at end of file
diff --git a/modeling/backbone/.__init__.py.swp b/modeling/backbone/.__init__.py.swp
new file mode 100644
index 0000000..e418868
Binary files /dev/null and b/modeling/backbone/.__init__.py.swp differ
diff --git a/modeling/backbone/.mobilenet.py.swp b/modeling/backbone/.mobilenet.py.swp
new file mode 100644
index 0000000..1bb895e
Binary files /dev/null and b/modeling/backbone/.mobilenet.py.swp differ
diff --git a/modeling/backbone/__init__.py b/modeling/backbone/__init__.py
new file mode 100644
index 0000000..5864c11
--- /dev/null
+++ b/modeling/backbone/__init__.py
@@ -0,0 +1,13 @@
+from modeling.backbone import resnet, xception, drn, mobilenet
+
+def build_backbone(backbone, output_stride, BatchNorm):
+    if backbone == 'resnet':
+        return resnet.ResNet101(output_stride, BatchNorm)
+    elif backbone == 'xception':
+        return xception.AlignedXception(output_stride, BatchNorm)
+    elif backbone == 'drn':
+        return drn.drn_d_54(BatchNorm)
+    elif backbone == 'mobilenet':
+        return mobilenet.MobileNetV2(output_stride, BatchNorm)
+    else:
+        raise NotImplementedError
diff --git a/modeling/backbone/drn.py b/modeling/backbone/drn.py
new file mode 100644
index 0000000..d07b577
--- /dev/null
+++ b/modeling/backbone/drn.py
@@ -0,0 +1,402 @@
+import torch.nn as nn
+import math
+import torch.utils.model_zoo as model_zoo
+from modeling.sync_batchnorm.batchnorm import SynchronizedBatchNorm2d
+
+webroot = 'https://tigress-web.princeton.edu/~fy/drn/models/'
+
+model_urls = {
+    'resnet50': 'https://download.pytorch.org/models/resnet50-19c8e357.pth',
+    'drn-c-26': webroot + 'drn_c_26-ddedf421.pth',
+    'drn-c-42': webroot + 'drn_c_42-9d336e8c.pth',
+    'drn-c-58': webroot + 'drn_c_58-0a53a92c.pth',
+    'drn-d-22': webroot + 'drn_d_22-4bd2f8ea.pth',
+    'drn-d-38': webroot + 'drn_d_38-eebb45f0.pth',
+    'drn-d-54': webroot + 'drn_d_54-0e0534ff.pth',
+    'drn-d-105': webroot + 'drn_d_105-12b40979.pth'
+}
+
+
+def conv3x3(in_planes, out_planes, stride=1, padding=1, dilation=1):
+    return nn.Conv2d(in_planes, out_planes, kernel_size=3, stride=stride,
+                     padding=padding, bias=False, dilation=dilation)
+
+
+class BasicBlock(nn.Module):
+    expansion = 1
+
+    def __init__(self, inplanes, planes, stride=1, downsample=None,
+                 dilation=(1, 1), residual=True, BatchNorm=None):
+        super(BasicBlock, self).__init__()
+        self.conv1 = conv3x3(inplanes, planes, stride,
+                             padding=dilation[0], dilation=dilation[0])
+        self.bn1 = BatchNorm(planes)
+        self.relu = nn.ReLU(inplace=True)
+        self.conv2 = conv3x3(planes, planes,
+                             padding=dilation[1], dilation=dilation[1])
+        self.bn2 = BatchNorm(planes)
+        self.downsample = downsample
+        self.stride = stride
+        self.residual = residual
+
+    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)
+        if self.residual:
+            out += residual
+        out = self.relu(out)
+
+        return out
+
+
+class Bottleneck(nn.Module):
+    expansion = 4
+
+    def __init__(self, inplanes, planes, stride=1, downsample=None,
+                 dilation=(1, 1), residual=True, BatchNorm=None):
+        super(Bottleneck, self).__init__()
+        self.conv1 = nn.Conv2d(inplanes, planes, kernel_size=1, bias=False)
+        self.bn1 = BatchNorm(planes)
+        self.conv2 = nn.Conv2d(planes, planes, kernel_size=3, stride=stride,
+                               padding=dilation[1], bias=False,
+                               dilation=dilation[1])
+        self.bn2 = BatchNorm(planes)
+        self.conv3 = nn.Conv2d(planes, planes * 4, kernel_size=1, bias=False)
+        self.bn3 = BatchNorm(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 DRN(nn.Module):
+
+    def __init__(self, block, layers, arch='D',
+                 channels=(16, 32, 64, 128, 256, 512, 512, 512),
+                 BatchNorm=None):
+        super(DRN, self).__init__()
+        self.inplanes = channels[0]
+        self.out_dim = channels[-1]
+        self.arch = arch
+
+        if arch == 'C':
+            self.conv1 = nn.Conv2d(3, channels[0], kernel_size=7, stride=1,
+                                   padding=3, bias=False)
+            self.bn1 = BatchNorm(channels[0])
+            self.relu = nn.ReLU(inplace=True)
+
+            self.layer1 = self._make_layer(
+                BasicBlock, channels[0], layers[0], stride=1, BatchNorm=BatchNorm)
+            self.layer2 = self._make_layer(
+                BasicBlock, channels[1], layers[1], stride=2, BatchNorm=BatchNorm)
+
+        elif arch == 'D':
+            self.layer0 = nn.Sequential(
+                nn.Conv2d(3, channels[0], kernel_size=7, stride=1, padding=3,
+                          bias=False),
+                BatchNorm(channels[0]),
+                nn.ReLU(inplace=True)
+            )
+
+            self.layer1 = self._make_conv_layers(
+                channels[0], layers[0], stride=1, BatchNorm=BatchNorm)
+            self.layer2 = self._make_conv_layers(
+                channels[1], layers[1], stride=2, BatchNorm=BatchNorm)
+
+        self.layer3 = self._make_layer(block, channels[2], layers[2], stride=2, BatchNorm=BatchNorm)
+        self.layer4 = self._make_layer(block, channels[3], layers[3], stride=2, BatchNorm=BatchNorm)
+        self.layer5 = self._make_layer(block, channels[4], layers[4],
+                                       dilation=2, new_level=False, BatchNorm=BatchNorm)
+        self.layer6 = None if layers[5] == 0 else \
+            self._make_layer(block, channels[5], layers[5], dilation=4,
+                             new_level=False, BatchNorm=BatchNorm)
+
+        if arch == 'C':
+            self.layer7 = None if layers[6] == 0 else \
+                self._make_layer(BasicBlock, channels[6], layers[6], dilation=2,
+                                 new_level=False, residual=False, BatchNorm=BatchNorm)
+            self.layer8 = None if layers[7] == 0 else \
+                self._make_layer(BasicBlock, channels[7], layers[7], dilation=1,
+                                 new_level=False, residual=False, BatchNorm=BatchNorm)
+        elif arch == 'D':
+            self.layer7 = None if layers[6] == 0 else \
+                self._make_conv_layers(channels[6], layers[6], dilation=2, BatchNorm=BatchNorm)
+            self.layer8 = None if layers[7] == 0 else \
+                self._make_conv_layers(channels[7], layers[7], dilation=1, BatchNorm=BatchNorm)
+
+        self._init_weight()
+
+    def _init_weight(self):
+        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, SynchronizedBatchNorm2d):
+                m.weight.data.fill_(1)
+                m.bias.data.zero_()
+            elif isinstance(m, nn.BatchNorm2d):
+                m.weight.data.fill_(1)
+                m.bias.data.zero_()
+
+
+    def _make_layer(self, block, planes, blocks, stride=1, dilation=1,
+                    new_level=True, residual=True, BatchNorm=None):
+        assert dilation == 1 or dilation % 2 == 0
+        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),
+                BatchNorm(planes * block.expansion),
+            )
+
+        layers = list()
+        layers.append(block(
+            self.inplanes, planes, stride, downsample,
+            dilation=(1, 1) if dilation == 1 else (
+                dilation // 2 if new_level else dilation, dilation),
+            residual=residual, BatchNorm=BatchNorm))
+        self.inplanes = planes * block.expansion
+        for i in range(1, blocks):
+            layers.append(block(self.inplanes, planes, residual=residual,
+                                dilation=(dilation, dilation), BatchNorm=BatchNorm))
+
+        return nn.Sequential(*layers)
+
+    def _make_conv_layers(self, channels, convs, stride=1, dilation=1, BatchNorm=None):
+        modules = []
+        for i in range(convs):
+            modules.extend([
+                nn.Conv2d(self.inplanes, channels, kernel_size=3,
+                          stride=stride if i == 0 else 1,
+                          padding=dilation, bias=False, dilation=dilation),
+                BatchNorm(channels),
+                nn.ReLU(inplace=True)])
+            self.inplanes = channels
+        return nn.Sequential(*modules)
+
+    def forward(self, x):
+        if self.arch == 'C':
+            x = self.conv1(x)
+            x = self.bn1(x)
+            x = self.relu(x)
+        elif self.arch == 'D':
+            x = self.layer0(x)
+
+        x = self.layer1(x)
+        x = self.layer2(x)
+
+        x = self.layer3(x)
+        low_level_feat = x
+
+        x = self.layer4(x)
+        x = self.layer5(x)
+
+        if self.layer6 is not None:
+            x = self.layer6(x)
+
+        if self.layer7 is not None:
+            x = self.layer7(x)
+
+        if self.layer8 is not None:
+            x = self.layer8(x)
+
+        return x, low_level_feat
+
+
+class DRN_A(nn.Module):
+
+    def __init__(self, block, layers, BatchNorm=None):
+        self.inplanes = 64
+        super(DRN_A, self).__init__()
+        self.out_dim = 512 * block.expansion
+        self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3,
+                               bias=False)
+        self.bn1 = BatchNorm(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], BatchNorm=BatchNorm)
+        self.layer2 = self._make_layer(block, 128, layers[1], stride=2, BatchNorm=BatchNorm)
+        self.layer3 = self._make_layer(block, 256, layers[2], stride=1,
+                                       dilation=2, BatchNorm=BatchNorm)
+        self.layer4 = self._make_layer(block, 512, layers[3], stride=1,
+                                       dilation=4, BatchNorm=BatchNorm)
+
+        self._init_weight()
+
+    def _init_weight(self):
+        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, SynchronizedBatchNorm2d):
+                m.weight.data.fill_(1)
+                m.bias.data.zero_()
+            elif isinstance(m, nn.BatchNorm2d):
+                m.weight.data.fill_(1)
+                m.bias.data.zero_()
+
+    def _make_layer(self, block, planes, blocks, stride=1, dilation=1, BatchNorm=None):
+        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),
+                BatchNorm(planes * block.expansion),
+            )
+
+        layers = []
+        layers.append(block(self.inplanes, planes, stride, downsample, BatchNorm=BatchNorm))
+        self.inplanes = planes * block.expansion
+        for i in range(1, blocks):
+            layers.append(block(self.inplanes, planes,
+                                dilation=(dilation, dilation, ), BatchNorm=BatchNorm))
+
+        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)
+
+        return x
+
+def drn_a_50(BatchNorm, pretrained=True):
+    model = DRN_A(Bottleneck, [3, 4, 6, 3], BatchNorm=BatchNorm)
+    if pretrained:
+        model.load_state_dict(model_zoo.load_url(model_urls['resnet50']))
+    return model
+
+
+def drn_c_26(BatchNorm, pretrained=True):
+    model = DRN(BasicBlock, [1, 1, 2, 2, 2, 2, 1, 1], arch='C', BatchNorm=BatchNorm)
+    if pretrained:
+        pretrained = model_zoo.load_url(model_urls['drn-c-26'])
+        del pretrained['fc.weight']
+        del pretrained['fc.bias']
+        model.load_state_dict(pretrained)
+    return model
+
+
+def drn_c_42(BatchNorm, pretrained=True):
+    model = DRN(BasicBlock, [1, 1, 3, 4, 6, 3, 1, 1], arch='C', BatchNorm=BatchNorm)
+    if pretrained:
+        pretrained = model_zoo.load_url(model_urls['drn-c-42'])
+        del pretrained['fc.weight']
+        del pretrained['fc.bias']
+        model.load_state_dict(pretrained)
+    return model
+
+
+def drn_c_58(BatchNorm, pretrained=True):
+    model = DRN(Bottleneck, [1, 1, 3, 4, 6, 3, 1, 1], arch='C', BatchNorm=BatchNorm)
+    if pretrained:
+        pretrained = model_zoo.load_url(model_urls['drn-c-58'])
+        del pretrained['fc.weight']
+        del pretrained['fc.bias']
+        model.load_state_dict(pretrained)
+    return model
+
+
+def drn_d_22(BatchNorm, pretrained=True):
+    model = DRN(BasicBlock, [1, 1, 2, 2, 2, 2, 1, 1], arch='D', BatchNorm=BatchNorm)
+    if pretrained:
+        pretrained = model_zoo.load_url(model_urls['drn-d-22'])
+        del pretrained['fc.weight']
+        del pretrained['fc.bias']
+        model.load_state_dict(pretrained)
+    return model
+
+
+def drn_d_24(BatchNorm, pretrained=True):
+    model = DRN(BasicBlock, [1, 1, 2, 2, 2, 2, 2, 2], arch='D', BatchNorm=BatchNorm)
+    if pretrained:
+        pretrained = model_zoo.load_url(model_urls['drn-d-24'])
+        del pretrained['fc.weight']
+        del pretrained['fc.bias']
+        model.load_state_dict(pretrained)
+    return model
+
+
+def drn_d_38(BatchNorm, pretrained=True):
+    model = DRN(BasicBlock, [1, 1, 3, 4, 6, 3, 1, 1], arch='D', BatchNorm=BatchNorm)
+    if pretrained:
+        pretrained = model_zoo.load_url(model_urls['drn-d-38'])
+        del pretrained['fc.weight']
+        del pretrained['fc.bias']
+        model.load_state_dict(pretrained)
+    return model
+
+
+def drn_d_40(BatchNorm, pretrained=True):
+    model = DRN(BasicBlock, [1, 1, 3, 4, 6, 3, 2, 2], arch='D', BatchNorm=BatchNorm)
+    if pretrained:
+        pretrained = model_zoo.load_url(model_urls['drn-d-40'])
+        del pretrained['fc.weight']
+        del pretrained['fc.bias']
+        model.load_state_dict(pretrained)
+    return model
+
+
+def drn_d_54(BatchNorm, pretrained=True):
+    model = DRN(Bottleneck, [1, 1, 3, 4, 6, 3, 1, 1], arch='D', BatchNorm=BatchNorm)
+    if pretrained:
+        pretrained = model_zoo.load_url(model_urls['drn-d-54'])
+        del pretrained['fc.weight']
+        del pretrained['fc.bias']
+        model.load_state_dict(pretrained)
+    return model
+
+
+def drn_d_105(BatchNorm, pretrained=True):
+    model = DRN(Bottleneck, [1, 1, 3, 4, 23, 3, 1, 1], arch='D', BatchNorm=BatchNorm)
+    if pretrained:
+        pretrained = model_zoo.load_url(model_urls['drn-d-105'])
+        del pretrained['fc.weight']
+        del pretrained['fc.bias']
+        model.load_state_dict(pretrained)
+    return model
+
+if __name__ == "__main__":
+    import torch
+    model = drn_a_50(BatchNorm=nn.BatchNorm2d, pretrained=True)
+    input = torch.rand(1, 3, 512, 512)
+    output, low_level_feat = model(input)
+    print(output.size())
+    print(low_level_feat.size())
diff --git a/modeling/backbone/mobilenet.py b/modeling/backbone/mobilenet.py
new file mode 100644
index 0000000..95410e9
--- /dev/null
+++ b/modeling/backbone/mobilenet.py
@@ -0,0 +1,156 @@
+import torch
+import torch.nn.functional as F
+import torch.nn as nn
+import math
+from modeling.sync_batchnorm.batchnorm import SynchronizedBatchNorm2d
+import torch.utils.model_zoo as model_zoo
+
+def conv_bn(inp, oup, stride, BatchNorm):
+    return nn.Sequential(
+        nn.Conv2d(inp, oup, 3, stride, 1, bias=False), #in_ch, out_ch, kernel, stride, pad
+        BatchNorm(oup),
+        nn.ReLU6(inplace=True)
+    )
+
+
+def fixed_padding(inputs, kernel_size, dilation):
+    kernel_size_effective = kernel_size + (kernel_size - 1) * (dilation - 1)
+    pad_total = kernel_size_effective - 1
+    pad_beg = pad_total // 2
+    pad_end = pad_total - pad_beg
+    padded_inputs = F.pad(inputs, (pad_beg, pad_end, pad_beg, pad_end))
+    return padded_inputs
+
+
+class InvertedResidual(nn.Module):
+    def __init__(self, inp, oup, stride, dilation, expand_ratio, BatchNorm):
+        super(InvertedResidual, self).__init__()
+        self.stride = stride
+        assert stride in [1, 2]
+
+        hidden_dim = round(inp * expand_ratio)
+        self.use_res_connect = self.stride == 1 and inp == oup
+        self.kernel_size = 3
+        self.dilation = dilation
+
+        if expand_ratio == 1:
+            self.conv = nn.Sequential(
+                # dw
+                nn.Conv2d(hidden_dim, hidden_dim, 3, stride, 0, dilation, groups=hidden_dim, bias=False),
+                BatchNorm(hidden_dim),
+                nn.ReLU6(inplace=True),
+                # pw-linear
+                nn.Conv2d(hidden_dim, oup, 1, 1, 0, 1, 1, bias=False),
+                BatchNorm(oup),
+            )
+        else:
+            self.conv = nn.Sequential(
+                # pw
+                nn.Conv2d(inp, hidden_dim, 1, 1, 0, 1, bias=False),
+                BatchNorm(hidden_dim),
+                nn.ReLU6(inplace=True),
+                # dw
+                nn.Conv2d(hidden_dim, hidden_dim, 3, stride, 0, dilation, groups=hidden_dim, bias=False),
+                BatchNorm(hidden_dim),
+                nn.ReLU6(inplace=True),
+                # pw-linear
+                nn.Conv2d(hidden_dim, oup, 1, 1, 0, 1, bias=False),
+                BatchNorm(oup),
+            )
+
+    def forward(self, x):
+        x_pad = fixed_padding(x, self.kernel_size, dilation=self.dilation)
+        if self.use_res_connect:
+            x = x + self.conv(x_pad)
+        else:
+            x = self.conv(x_pad)
+        return x
+
+
+class MobileNetV2(nn.Module):
+    def __init__(self, output_stride=8, BatchNorm=None, width_mult=1., pretrained=True):
+        #ltt add
+        pretrained = False
+        super(MobileNetV2, self).__init__()
+        block = InvertedResidual
+        input_channel = 32 #????P的32, 网络的第一层由conv_bn 的channel决定
+        current_stride = 1
+        rate = 1
+        interverted_residual_setting = [
+            # t, c, n, s
+            [1, 16, 1, 1],
+            [6, 24, 2, 2],
+            [6, 32, 3, 2],
+            [6, 64, 4, 2],
+            [6, 96, 3, 1],
+            [6, 160, 3, 2],
+            [6, 320, 1, 1],
+        ]
+
+        # building first layer
+        input_channel = int(input_channel * width_mult)
+        self.features = [conv_bn(1, input_channel, 2, BatchNorm)] #input_channel成为输出层
+        print("first layer success")
+        current_stride *= 2
+        # building inverted residual blocks
+        for t, c, n, s in interverted_residual_setting:
+            if current_stride == output_stride: #达到输出的步长就不变了
+                stride = 1
+                dilation = rate #空洞率
+                rate *= s #s是空洞倍率
+            else:
+                stride = s
+                dilation = 1
+                current_stride *= s
+            output_channel = int(c * width_mult)
+            for i in range(n):
+                if i == 0:
+                    self.features.append(block(input_channel, output_channel, stride, dilation, t, BatchNorm))
+                else:
+                    self.features.append(block(input_channel, output_channel, 1, dilation, t, BatchNorm))
+                input_channel = output_channel
+        self.features = nn.Sequential(*self.features)
+        self._initialize_weights()
+        print("finish init")
+        if pretrained:
+            self._load_pretrained_model()
+        self.low_level_features = self.features[0:4]
+        self.high_level_features = self.features[4:]
+
+    def forward(self, x):
+        print("before low features")
+        x = torch.tensor(x, dtype=torch.float32)
+        low_level_feat = self.low_level_features(x)
+        print("after low feature")
+        x = self.high_level_features(low_level_feat)
+        return x, low_level_feat
+
+    def _load_pretrained_model(self):
+        pretrain_dict = model_zoo.load_url('http://jeff95.me/models/mobilenet_v2-6a65762b.pth')
+        model_dict = {}
+        state_dict = self.state_dict()
+        for k, v in pretrain_dict.items():
+            if k in state_dict:
+                model_dict[k] = v
+        state_dict.update(model_dict)
+        self.load_state_dict(state_dict)
+
+    def _initialize_weights(self):
+        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))
+                torch.nn.init.kaiming_normal_(m.weight)
+            elif isinstance(m, SynchronizedBatchNorm2d):
+                m.weight.data.fill_(1)
+                m.bias.data.zero_()
+            elif isinstance(m, nn.BatchNorm2d):
+                m.weight.data.fill_(1)
+                m.bias.data.zero_()
+
+if __name__ == "__main__":
+    input = torch.rand(1, 3, 512, 512)
+    model = MobileNetV2(output_stride=16, BatchNorm=nn.BatchNorm2d)
+    output, low_level_feat = model(input)
+    print(output.size())
+    print(low_level_feat.size())
diff --git a/modeling/backbone/resnet.py b/modeling/backbone/resnet.py
new file mode 100644
index 0000000..8eeac69
--- /dev/null
+++ b/modeling/backbone/resnet.py
@@ -0,0 +1,162 @@
+import math
+import torch.nn as nn
+import torch.utils.model_zoo as model_zoo
+from modeling.sync_batchnorm.batchnorm import SynchronizedBatchNorm2d
+
+class Bottleneck(nn.Module):
+    expansion = 4
+
+    def __init__(self, inplanes, planes, stride=1, dilation=1, downsample=None, BatchNorm=None):
+        super(Bottleneck, self).__init__()
+        self.conv1 = nn.Conv2d(inplanes, planes, kernel_size=1, bias=False)
+        self.bn1 = BatchNorm(planes)
+        self.conv2 = nn.Conv2d(planes, planes, kernel_size=3, stride=stride,
+                               dilation=dilation, padding=dilation, bias=False)
+        self.bn2 = BatchNorm(planes)
+        self.conv3 = nn.Conv2d(planes, planes * 4, kernel_size=1, bias=False)
+        self.bn3 = BatchNorm(planes * 4)
+        self.relu = nn.ReLU(inplace=True)
+        self.downsample = downsample
+        self.stride = stride
+        self.dilation = dilation
+
+    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, output_stride, BatchNorm, pretrained=True):
+        self.inplanes = 64
+        super(ResNet, self).__init__()
+        blocks = [1, 2, 4]
+        if output_stride == 16:
+            strides = [1, 2, 2, 1]
+            dilations = [1, 1, 1, 2]
+        elif output_stride == 8:
+            strides = [1, 2, 1, 1]
+            dilations = [1, 1, 2, 4]
+        else:
+            raise NotImplementedError
+
+        # Modules
+        self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3,
+                                bias=False)
+        self.bn1 = BatchNorm(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], stride=strides[0], dilation=dilations[0], BatchNorm=BatchNorm)
+        self.layer2 = self._make_layer(block, 128, layers[1], stride=strides[1], dilation=dilations[1], BatchNorm=BatchNorm)
+        self.layer3 = self._make_layer(block, 256, layers[2], stride=strides[2], dilation=dilations[2], BatchNorm=BatchNorm)
+        self.layer4 = self._make_MG_unit(block, 512, blocks=blocks, stride=strides[3], dilation=dilations[3], BatchNorm=BatchNorm)
+        # self.layer4 = self._make_layer(block, 512, layers[3], stride=strides[3], dilation=dilations[3], BatchNorm=BatchNorm)
+        self._init_weight()
+
+        if pretrained:
+            self._load_pretrained_model()
+
+    def _make_layer(self, block, planes, blocks, stride=1, dilation=1, BatchNorm=None):
+        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),
+                BatchNorm(planes * block.expansion),
+            )
+
+        layers = []
+        layers.append(block(self.inplanes, planes, stride, dilation, downsample, BatchNorm))
+        self.inplanes = planes * block.expansion
+        for i in range(1, blocks):
+            layers.append(block(self.inplanes, planes, dilation=dilation, BatchNorm=BatchNorm))
+
+        return nn.Sequential(*layers)
+
+    def _make_MG_unit(self, block, planes, blocks, stride=1, dilation=1, BatchNorm=None):
+        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),
+                BatchNorm(planes * block.expansion),
+            )
+
+        layers = []
+        layers.append(block(self.inplanes, planes, stride, dilation=blocks[0]*dilation,
+                            downsample=downsample, BatchNorm=BatchNorm))
+        self.inplanes = planes * block.expansion
+        for i in range(1, len(blocks)):
+            layers.append(block(self.inplanes, planes, stride=1,
+                                dilation=blocks[i]*dilation, BatchNorm=BatchNorm))
+
+        return nn.Sequential(*layers)
+
+    def forward(self, input):
+        x = self.conv1(input)
+        x = self.bn1(x)
+        x = self.relu(x)
+        x = self.maxpool(x)
+
+        x = self.layer1(x)
+        low_level_feat = x
+        x = self.layer2(x)
+        x = self.layer3(x)
+        x = self.layer4(x)
+        return x, low_level_feat
+
+    def _init_weight(self):
+        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, SynchronizedBatchNorm2d):
+                m.weight.data.fill_(1)
+                m.bias.data.zero_()
+            elif isinstance(m, nn.BatchNorm2d):
+                m.weight.data.fill_(1)
+                m.bias.data.zero_()
+
+    def _load_pretrained_model(self):
+        pretrain_dict = model_zoo.load_url('https://download.pytorch.org/models/resnet101-5d3b4d8f.pth')
+        model_dict = {}
+        state_dict = self.state_dict()
+        for k, v in pretrain_dict.items():
+            if k in state_dict:
+                model_dict[k] = v
+        state_dict.update(model_dict)
+        self.load_state_dict(state_dict)
+
+def ResNet101(output_stride, BatchNorm, pretrained=True):
+    """Constructs a ResNet-101 model.
+    Args:
+        pretrained (bool): If True, returns a model pre-trained on ImageNet
+    """
+    model = ResNet(Bottleneck, [3, 4, 23, 3], output_stride, BatchNorm, pretrained=pretrained)
+    return model
+
+if __name__ == "__main__":
+    import torch
+    model = ResNet101(BatchNorm=nn.BatchNorm2d, pretrained=True, output_stride=8)
+    input = torch.rand(1, 3, 512, 512)
+    output, low_level_feat = model(input)
+    print(output.size())
+    print(low_level_feat.size())
\ No newline at end of file
diff --git a/modeling/backbone/xception.py b/modeling/backbone/xception.py
new file mode 100644
index 0000000..4752367
--- /dev/null
+++ b/modeling/backbone/xception.py
@@ -0,0 +1,288 @@
+import math
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+import torch.utils.model_zoo as model_zoo
+from modeling.sync_batchnorm.batchnorm import SynchronizedBatchNorm2d
+
+def fixed_padding(inputs, kernel_size, dilation):
+    kernel_size_effective = kernel_size + (kernel_size - 1) * (dilation - 1)
+    pad_total = kernel_size_effective - 1
+    pad_beg = pad_total // 2
+    pad_end = pad_total - pad_beg
+    padded_inputs = F.pad(inputs, (pad_beg, pad_end, pad_beg, pad_end))
+    return padded_inputs
+
+
+class SeparableConv2d(nn.Module):
+    def __init__(self, inplanes, planes, kernel_size=3, stride=1, dilation=1, bias=False, BatchNorm=None):
+        super(SeparableConv2d, self).__init__()
+
+        self.conv1 = nn.Conv2d(inplanes, inplanes, kernel_size, stride, 0, dilation,
+                               groups=inplanes, bias=bias)
+        self.bn = BatchNorm(inplanes)
+        self.pointwise = nn.Conv2d(inplanes, planes, 1, 1, 0, 1, 1, bias=bias)
+
+    def forward(self, x):
+        x = fixed_padding(x, self.conv1.kernel_size[0], dilation=self.conv1.dilation[0])
+        x = self.conv1(x)
+        x = self.bn(x)
+        x = self.pointwise(x)
+        return x
+
+
+class Block(nn.Module):
+    def __init__(self, inplanes, planes, reps, stride=1, dilation=1, BatchNorm=None,
+                 start_with_relu=True, grow_first=True, is_last=False):
+        super(Block, self).__init__()
+
+        if planes != inplanes or stride != 1:
+            self.skip = nn.Conv2d(inplanes, planes, 1, stride=stride, bias=False)
+            self.skipbn = BatchNorm(planes)
+        else:
+            self.skip = None
+
+        self.relu = nn.ReLU(inplace=True)
+        rep = []
+
+        filters = inplanes
+        if grow_first:
+            rep.append(self.relu)
+            rep.append(SeparableConv2d(inplanes, planes, 3, 1, dilation, BatchNorm=BatchNorm))
+            rep.append(BatchNorm(planes))
+            filters = planes
+
+        for i in range(reps - 1):
+            rep.append(self.relu)
+            rep.append(SeparableConv2d(filters, filters, 3, 1, dilation, BatchNorm=BatchNorm))
+            rep.append(BatchNorm(filters))
+
+        if not grow_first:
+            rep.append(self.relu)
+            rep.append(SeparableConv2d(inplanes, planes, 3, 1, dilation, BatchNorm=BatchNorm))
+            rep.append(BatchNorm(planes))
+
+        if stride != 1:
+            rep.append(self.relu)
+            rep.append(SeparableConv2d(planes, planes, 3, 2, BatchNorm=BatchNorm))
+            rep.append(BatchNorm(planes))
+
+        if stride == 1 and is_last:
+            rep.append(self.relu)
+            rep.append(SeparableConv2d(planes, planes, 3, 1, BatchNorm=BatchNorm))
+            rep.append(BatchNorm(planes))
+
+        if not start_with_relu:
+            rep = rep[1:]
+
+        self.rep = nn.Sequential(*rep)
+
+    def forward(self, inp):
+        x = self.rep(inp)
+
+        if self.skip is not None:
+            skip = self.skip(inp)
+            skip = self.skipbn(skip)
+        else:
+            skip = inp
+
+        x = x + skip
+
+        return x
+
+
+class AlignedXception(nn.Module):
+    """
+    Modified Alighed Xception
+    """
+    def __init__(self, output_stride, BatchNorm,
+                 pretrained=True):
+        super(AlignedXception, self).__init__()
+
+        if output_stride == 16:
+            entry_block3_stride = 2
+            middle_block_dilation = 1
+            exit_block_dilations = (1, 2)
+        elif output_stride == 8:
+            entry_block3_stride = 1
+            middle_block_dilation = 2
+            exit_block_dilations = (2, 4)
+        else:
+            raise NotImplementedError
+
+
+        # Entry flow
+        self.conv1 = nn.Conv2d(3, 32, 3, stride=2, padding=1, bias=False)
+        self.bn1 = BatchNorm(32)
+        self.relu = nn.ReLU(inplace=True)
+
+        self.conv2 = nn.Conv2d(32, 64, 3, stride=1, padding=1, bias=False)
+        self.bn2 = BatchNorm(64)
+
+        self.block1 = Block(64, 128, reps=2, stride=2, BatchNorm=BatchNorm, start_with_relu=False)
+        self.block2 = Block(128, 256, reps=2, stride=2, BatchNorm=BatchNorm, start_with_relu=False,
+                            grow_first=True)
+        self.block3 = Block(256, 728, reps=2, stride=entry_block3_stride, BatchNorm=BatchNorm,
+                            start_with_relu=True, grow_first=True, is_last=True)
+
+        # Middle flow
+        self.block4  = Block(728, 728, reps=3, stride=1, dilation=middle_block_dilation,
+                             BatchNorm=BatchNorm, start_with_relu=True, grow_first=True)
+        self.block5  = Block(728, 728, reps=3, stride=1, dilation=middle_block_dilation,
+                             BatchNorm=BatchNorm, start_with_relu=True, grow_first=True)
+        self.block6  = Block(728, 728, reps=3, stride=1, dilation=middle_block_dilation,
+                             BatchNorm=BatchNorm, start_with_relu=True, grow_first=True)
+        self.block7  = Block(728, 728, reps=3, stride=1, dilation=middle_block_dilation,
+                             BatchNorm=BatchNorm, start_with_relu=True, grow_first=True)
+        self.block8  = Block(728, 728, reps=3, stride=1, dilation=middle_block_dilation,
+                             BatchNorm=BatchNorm, start_with_relu=True, grow_first=True)
+        self.block9  = Block(728, 728, reps=3, stride=1, dilation=middle_block_dilation,
+                             BatchNorm=BatchNorm, start_with_relu=True, grow_first=True)
+        self.block10 = Block(728, 728, reps=3, stride=1, dilation=middle_block_dilation,
+                             BatchNorm=BatchNorm, start_with_relu=True, grow_first=True)
+        self.block11 = Block(728, 728, reps=3, stride=1, dilation=middle_block_dilation,
+                             BatchNorm=BatchNorm, start_with_relu=True, grow_first=True)
+        self.block12 = Block(728, 728, reps=3, stride=1, dilation=middle_block_dilation,
+                             BatchNorm=BatchNorm, start_with_relu=True, grow_first=True)
+        self.block13 = Block(728, 728, reps=3, stride=1, dilation=middle_block_dilation,
+                             BatchNorm=BatchNorm, start_with_relu=True, grow_first=True)
+        self.block14 = Block(728, 728, reps=3, stride=1, dilation=middle_block_dilation,
+                             BatchNorm=BatchNorm, start_with_relu=True, grow_first=True)
+        self.block15 = Block(728, 728, reps=3, stride=1, dilation=middle_block_dilation,
+                             BatchNorm=BatchNorm, start_with_relu=True, grow_first=True)
+        self.block16 = Block(728, 728, reps=3, stride=1, dilation=middle_block_dilation,
+                             BatchNorm=BatchNorm, start_with_relu=True, grow_first=True)
+        self.block17 = Block(728, 728, reps=3, stride=1, dilation=middle_block_dilation,
+                             BatchNorm=BatchNorm, start_with_relu=True, grow_first=True)
+        self.block18 = Block(728, 728, reps=3, stride=1, dilation=middle_block_dilation,
+                             BatchNorm=BatchNorm, start_with_relu=True, grow_first=True)
+        self.block19 = Block(728, 728, reps=3, stride=1, dilation=middle_block_dilation,
+                             BatchNorm=BatchNorm, start_with_relu=True, grow_first=True)
+
+        # Exit flow
+        self.block20 = Block(728, 1024, reps=2, stride=1, dilation=exit_block_dilations[0],
+                             BatchNorm=BatchNorm, start_with_relu=True, grow_first=False, is_last=True)
+
+        self.conv3 = SeparableConv2d(1024, 1536, 3, stride=1, dilation=exit_block_dilations[1], BatchNorm=BatchNorm)
+        self.bn3 = BatchNorm(1536)
+
+        self.conv4 = SeparableConv2d(1536, 1536, 3, stride=1, dilation=exit_block_dilations[1], BatchNorm=BatchNorm)
+        self.bn4 = BatchNorm(1536)
+
+        self.conv5 = SeparableConv2d(1536, 2048, 3, stride=1, dilation=exit_block_dilations[1], BatchNorm=BatchNorm)
+        self.bn5 = BatchNorm(2048)
+
+        # Init weights
+        self._init_weight()
+
+        # Load pretrained model
+        if pretrained:
+            self._load_pretrained_model()
+
+    def forward(self, x):
+        # Entry flow
+        x = self.conv1(x)
+        x = self.bn1(x)
+        x = self.relu(x)
+
+        x = self.conv2(x)
+        x = self.bn2(x)
+        x = self.relu(x)
+
+        x = self.block1(x)
+        # add relu here
+        x = self.relu(x)
+        low_level_feat = x
+        x = self.block2(x)
+        x = self.block3(x)
+
+        # Middle flow
+        x = self.block4(x)
+        x = self.block5(x)
+        x = self.block6(x)
+        x = self.block7(x)
+        x = self.block8(x)
+        x = self.block9(x)
+        x = self.block10(x)
+        x = self.block11(x)
+        x = self.block12(x)
+        x = self.block13(x)
+        x = self.block14(x)
+        x = self.block15(x)
+        x = self.block16(x)
+        x = self.block17(x)
+        x = self.block18(x)
+        x = self.block19(x)
+
+        # Exit flow
+        x = self.block20(x)
+        x = self.relu(x)
+        x = self.conv3(x)
+        x = self.bn3(x)
+        x = self.relu(x)
+
+        x = self.conv4(x)
+        x = self.bn4(x)
+        x = self.relu(x)
+
+        x = self.conv5(x)
+        x = self.bn5(x)
+        x = self.relu(x)
+
+        return x, low_level_feat
+
+    def _init_weight(self):
+        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, SynchronizedBatchNorm2d):
+                m.weight.data.fill_(1)
+                m.bias.data.zero_()
+            elif isinstance(m, nn.BatchNorm2d):
+                m.weight.data.fill_(1)
+                m.bias.data.zero_()
+
+
+    def _load_pretrained_model(self):
+        pretrain_dict = model_zoo.load_url('http://data.lip6.fr/cadene/pretrainedmodels/xception-b5690688.pth')
+        model_dict = {}
+        state_dict = self.state_dict()
+
+        for k, v in pretrain_dict.items():
+            if k in model_dict:
+                if 'pointwise' in k:
+                    v = v.unsqueeze(-1).unsqueeze(-1)
+                if k.startswith('block11'):
+                    model_dict[k] = v
+                    model_dict[k.replace('block11', 'block12')] = v
+                    model_dict[k.replace('block11', 'block13')] = v
+                    model_dict[k.replace('block11', 'block14')] = v
+                    model_dict[k.replace('block11', 'block15')] = v
+                    model_dict[k.replace('block11', 'block16')] = v
+                    model_dict[k.replace('block11', 'block17')] = v
+                    model_dict[k.replace('block11', 'block18')] = v
+                    model_dict[k.replace('block11', 'block19')] = v
+                elif k.startswith('block12'):
+                    model_dict[k.replace('block12', 'block20')] = v
+                elif k.startswith('bn3'):
+                    model_dict[k] = v
+                    model_dict[k.replace('bn3', 'bn4')] = v
+                elif k.startswith('conv4'):
+                    model_dict[k.replace('conv4', 'conv5')] = v
+                elif k.startswith('bn4'):
+                    model_dict[k.replace('bn4', 'bn5')] = v
+                else:
+                    model_dict[k] = v
+        state_dict.update(model_dict)
+        self.load_state_dict(state_dict)
+
+
+
+if __name__ == "__main__":
+    import torch
+    model = AlignedXception(BatchNorm=nn.BatchNorm2d, pretrained=True, output_stride=16)
+    input = torch.rand(1, 3, 512, 512)
+    output, low_level_feat = model(input)
+    print(output.size())
+    print(low_level_feat.size())
\ No newline at end of file
diff --git a/modeling/decoder.py b/modeling/decoder.py
new file mode 100644
index 0000000..5ed41d0
--- /dev/null
+++ b/modeling/decoder.py
@@ -0,0 +1,57 @@
+import math
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from modeling.sync_batchnorm.batchnorm import SynchronizedBatchNorm2d
+
+class Decoder(nn.Module):
+    def __init__(self, num_classes, backbone, BatchNorm):
+        super(Decoder, self).__init__()
+        if backbone == 'resnet' or backbone == 'drn':
+            low_level_inplanes = 256
+        elif backbone == 'xception':
+            low_level_inplanes = 128
+        elif backbone == 'mobilenet':
+            low_level_inplanes = 24
+        else:
+            raise NotImplementedError
+
+        self.conv1 = nn.Conv2d(low_level_inplanes, 48, 1, bias=False)
+        self.bn1 = BatchNorm(48)
+        self.relu = nn.ReLU()
+        self.last_conv = nn.Sequential(nn.Conv2d(304, 256, kernel_size=3, stride=1, padding=1, bias=False),
+                                       BatchNorm(256),
+                                       nn.ReLU(),
+                                       nn.Dropout(0.5),
+                                       nn.Conv2d(256, 256, kernel_size=3, stride=1, padding=1, bias=False),
+                                       BatchNorm(256),
+                                       nn.ReLU(),
+                                       nn.Dropout(0.1),
+                                       nn.Conv2d(256, num_classes, kernel_size=1, stride=1))
+        self._init_weight()
+
+
+    def forward(self, x, low_level_feat):
+        low_level_feat = self.conv1(low_level_feat)
+        low_level_feat = self.bn1(low_level_feat)
+        low_level_feat = self.relu(low_level_feat)
+
+        x = F.interpolate(x, size=low_level_feat.size()[2:], mode='bilinear', align_corners=True)
+        x = torch.cat((x, low_level_feat), dim=1)
+        x = self.last_conv(x)
+
+        return x
+
+    def _init_weight(self):
+        for m in self.modules():
+            if isinstance(m, nn.Conv2d):
+                torch.nn.init.kaiming_normal_(m.weight)
+            elif isinstance(m, SynchronizedBatchNorm2d):
+                m.weight.data.fill_(1)
+                m.bias.data.zero_()
+            elif isinstance(m, nn.BatchNorm2d):
+                m.weight.data.fill_(1)
+                m.bias.data.zero_()
+
+def build_decoder(num_classes, backbone, BatchNorm):
+    return Decoder(num_classes, backbone, BatchNorm)
\ No newline at end of file
diff --git a/modeling/sync_batchnorm/__init__.py b/modeling/sync_batchnorm/__init__.py
new file mode 100644
index 0000000..540118f
--- /dev/null
+++ b/modeling/sync_batchnorm/__init__.py
@@ -0,0 +1,12 @@
+# -*- coding: utf-8 -*-
+# File   : __init__.py
+# Author : Jiayuan Mao
+# Email  : maojiayuan@gmail.com
+# Date   : 27/01/2018
+#
+# This file is part of Synchronized-BatchNorm-PyTorch.
+# https://github.com/vacancy/Synchronized-BatchNorm-PyTorch
+# Distributed under MIT License.
+
+from .batchnorm import SynchronizedBatchNorm1d, SynchronizedBatchNorm2d, SynchronizedBatchNorm3d
+from .replicate import DataParallelWithCallback, patch_replication_callback
\ No newline at end of file
diff --git a/modeling/sync_batchnorm/batchnorm.py b/modeling/sync_batchnorm/batchnorm.py
new file mode 100644
index 0000000..aa9dd37
--- /dev/null
+++ b/modeling/sync_batchnorm/batchnorm.py
@@ -0,0 +1,282 @@
+# -*- coding: utf-8 -*-
+# File   : batchnorm.py
+# Author : Jiayuan Mao
+# Email  : maojiayuan@gmail.com
+# Date   : 27/01/2018
+#
+# This file is part of Synchronized-BatchNorm-PyTorch.
+# https://github.com/vacancy/Synchronized-BatchNorm-PyTorch
+# Distributed under MIT License.
+
+import collections
+
+import torch
+import torch.nn.functional as F
+
+from torch.nn.modules.batchnorm import _BatchNorm
+from torch.nn.parallel._functions import ReduceAddCoalesced, Broadcast
+
+from .comm import SyncMaster
+
+__all__ = ['SynchronizedBatchNorm1d', 'SynchronizedBatchNorm2d', 'SynchronizedBatchNorm3d']
+
+
+def _sum_ft(tensor):
+    """sum over the first and last dimention"""
+    return tensor.sum(dim=0).sum(dim=-1)
+
+
+def _unsqueeze_ft(tensor):
+    """add new dementions at the front and the tail"""
+    return tensor.unsqueeze(0).unsqueeze(-1)
+
+
+_ChildMessage = collections.namedtuple('_ChildMessage', ['sum', 'ssum', 'sum_size'])
+_MasterMessage = collections.namedtuple('_MasterMessage', ['sum', 'inv_std'])
+
+
+class _SynchronizedBatchNorm(_BatchNorm):
+    def __init__(self, num_features, eps=1e-5, momentum=0.1, affine=True):
+        super(_SynchronizedBatchNorm, self).__init__(num_features, eps=eps, momentum=momentum, affine=affine)
+
+        self._sync_master = SyncMaster(self._data_parallel_master)
+
+        self._is_parallel = False
+        self._parallel_id = None
+        self._slave_pipe = None
+
+    def forward(self, input):
+        # If it is not parallel computation or is in evaluation mode, use PyTorch's implementation.
+        if not (self._is_parallel and self.training):
+            return F.batch_norm(
+                input, self.running_mean, self.running_var, self.weight, self.bias,
+                self.training, self.momentum, self.eps)
+
+        # Resize the input to (B, C, -1).
+        input_shape = input.size()
+        input = input.view(input.size(0), self.num_features, -1)
+
+        # Compute the sum and square-sum.
+        sum_size = input.size(0) * input.size(2)
+        input_sum = _sum_ft(input)
+        input_ssum = _sum_ft(input ** 2)
+
+        # Reduce-and-broadcast the statistics.
+        if self._parallel_id == 0:
+            mean, inv_std = self._sync_master.run_master(_ChildMessage(input_sum, input_ssum, sum_size))
+        else:
+            mean, inv_std = self._slave_pipe.run_slave(_ChildMessage(input_sum, input_ssum, sum_size))
+
+        # Compute the output.
+        if self.affine:
+            # MJY:: Fuse the multiplication for speed.
+            output = (input - _unsqueeze_ft(mean)) * _unsqueeze_ft(inv_std * self.weight) + _unsqueeze_ft(self.bias)
+        else:
+            output = (input - _unsqueeze_ft(mean)) * _unsqueeze_ft(inv_std)
+
+        # Reshape it.
+        return output.view(input_shape)
+
+    def __data_parallel_replicate__(self, ctx, copy_id):
+        self._is_parallel = True
+        self._parallel_id = copy_id
+
+        # parallel_id == 0 means master device.
+        if self._parallel_id == 0:
+            ctx.sync_master = self._sync_master
+        else:
+            self._slave_pipe = ctx.sync_master.register_slave(copy_id)
+
+    def _data_parallel_master(self, intermediates):
+        """Reduce the sum and square-sum, compute the statistics, and broadcast it."""
+
+        # Always using same "device order" makes the ReduceAdd operation faster.
+        # Thanks to:: Tete Xiao (http://tetexiao.com/)
+        intermediates = sorted(intermediates, key=lambda i: i[1].sum.get_device())
+
+        to_reduce = [i[1][:2] for i in intermediates]
+        to_reduce = [j for i in to_reduce for j in i]  # flatten
+        target_gpus = [i[1].sum.get_device() for i in intermediates]
+
+        sum_size = sum([i[1].sum_size for i in intermediates])
+        sum_, ssum = ReduceAddCoalesced.apply(target_gpus[0], 2, *to_reduce)
+        mean, inv_std = self._compute_mean_std(sum_, ssum, sum_size)
+
+        broadcasted = Broadcast.apply(target_gpus, mean, inv_std)
+
+        outputs = []
+        for i, rec in enumerate(intermediates):
+            outputs.append((rec[0], _MasterMessage(*broadcasted[i * 2:i * 2 + 2])))
+
+        return outputs
+
+    def _compute_mean_std(self, sum_, ssum, size):
+        """Compute the mean and standard-deviation with sum and square-sum. This method
+        also maintains the moving average on the master device."""
+        assert size > 1, 'BatchNorm computes unbiased standard-deviation, which requires size > 1.'
+        mean = sum_ / size
+        sumvar = ssum - sum_ * mean
+        unbias_var = sumvar / (size - 1)
+        bias_var = sumvar / size
+
+        self.running_mean = (1 - self.momentum) * self.running_mean + self.momentum * mean.data
+        self.running_var = (1 - self.momentum) * self.running_var + self.momentum * unbias_var.data
+
+        return mean, bias_var.clamp(self.eps) ** -0.5
+
+
+class SynchronizedBatchNorm1d(_SynchronizedBatchNorm):
+    r"""Applies Synchronized Batch Normalization over a 2d or 3d input that is seen as a
+    mini-batch.
+    .. math::
+        y = \frac{x - mean[x]}{ \sqrt{Var[x] + \epsilon}} * gamma + beta
+    This module differs from the built-in PyTorch BatchNorm1d as the mean and
+    standard-deviation are reduced across all devices during training.
+    For example, when one uses `nn.DataParallel` to wrap the network during
+    training, PyTorch's implementation normalize the tensor on each device using
+    the statistics only on that device, which accelerated the computation and
+    is also easy to implement, but the statistics might be inaccurate.
+    Instead, in this synchronized version, the statistics will be computed
+    over all training samples distributed on multiple devices.
+
+    Note that, for one-GPU or CPU-only case, this module behaves exactly same
+    as the built-in PyTorch implementation.
+    The mean and standard-deviation are calculated per-dimension over
+    the mini-batches and gamma and beta are learnable parameter vectors
+    of size C (where C is the input size).
+    During training, this layer keeps a running estimate of its computed mean
+    and variance. The running sum is kept with a default momentum of 0.1.
+    During evaluation, this running mean/variance is used for normalization.
+    Because the BatchNorm is done over the `C` dimension, computing statistics
+    on `(N, L)` slices, it's common terminology to call this Temporal BatchNorm
+    Args:
+        num_features: num_features from an expected input of size
+            `batch_size x num_features [x width]`
+        eps: a value added to the denominator for numerical stability.
+            Default: 1e-5
+        momentum: the value used for the running_mean and running_var
+            computation. Default: 0.1
+        affine: a boolean value that when set to ``True``, gives the layer learnable
+            affine parameters. Default: ``True``
+    Shape:
+        - Input: :math:`(N, C)` or :math:`(N, C, L)`
+        - Output: :math:`(N, C)` or :math:`(N, C, L)` (same shape as input)
+    Examples:
+        >>> # With Learnable Parameters
+        >>> m = SynchronizedBatchNorm1d(100)
+        >>> # Without Learnable Parameters
+        >>> m = SynchronizedBatchNorm1d(100, affine=False)
+        >>> input = torch.autograd.Variable(torch.randn(20, 100))
+        >>> output = m(input)
+    """
+
+    def _check_input_dim(self, input):
+        if input.dim() != 2 and input.dim() != 3:
+            raise ValueError('expected 2D or 3D input (got {}D input)'
+                             .format(input.dim()))
+        super(SynchronizedBatchNorm1d, self)._check_input_dim(input)
+
+
+class SynchronizedBatchNorm2d(_SynchronizedBatchNorm):
+    r"""Applies Batch Normalization over a 4d input that is seen as a mini-batch
+    of 3d inputs
+    .. math::
+        y = \frac{x - mean[x]}{ \sqrt{Var[x] + \epsilon}} * gamma + beta
+    This module differs from the built-in PyTorch BatchNorm2d as the mean and
+    standard-deviation are reduced across all devices during training.
+    For example, when one uses `nn.DataParallel` to wrap the network during
+    training, PyTorch's implementation normalize the tensor on each device using
+    the statistics only on that device, which accelerated the computation and
+    is also easy to implement, but the statistics might be inaccurate.
+    Instead, in this synchronized version, the statistics will be computed
+    over all training samples distributed on multiple devices.
+
+    Note that, for one-GPU or CPU-only case, this module behaves exactly same
+    as the built-in PyTorch implementation.
+    The mean and standard-deviation are calculated per-dimension over
+    the mini-batches and gamma and beta are learnable parameter vectors
+    of size C (where C is the input size).
+    During training, this layer keeps a running estimate of its computed mean
+    and variance. The running sum is kept with a default momentum of 0.1.
+    During evaluation, this running mean/variance is used for normalization.
+    Because the BatchNorm is done over the `C` dimension, computing statistics
+    on `(N, H, W)` slices, it's common terminology to call this Spatial BatchNorm
+    Args:
+        num_features: num_features from an expected input of
+            size batch_size x num_features x height x width
+        eps: a value added to the denominator for numerical stability.
+            Default: 1e-5
+        momentum: the value used for the running_mean and running_var
+            computation. Default: 0.1
+        affine: a boolean value that when set to ``True``, gives the layer learnable
+            affine parameters. Default: ``True``
+    Shape:
+        - Input: :math:`(N, C, H, W)`
+        - Output: :math:`(N, C, H, W)` (same shape as input)
+    Examples:
+        >>> # With Learnable Parameters
+        >>> m = SynchronizedBatchNorm2d(100)
+        >>> # Without Learnable Parameters
+        >>> m = SynchronizedBatchNorm2d(100, affine=False)
+        >>> input = torch.autograd.Variable(torch.randn(20, 100, 35, 45))
+        >>> output = m(input)
+    """
+
+    def _check_input_dim(self, input):
+        if input.dim() != 4:
+            raise ValueError('expected 4D input (got {}D input)'
+                             .format(input.dim()))
+        super(SynchronizedBatchNorm2d, self)._check_input_dim(input)
+
+
+class SynchronizedBatchNorm3d(_SynchronizedBatchNorm):
+    r"""Applies Batch Normalization over a 5d input that is seen as a mini-batch
+    of 4d inputs
+    .. math::
+        y = \frac{x - mean[x]}{ \sqrt{Var[x] + \epsilon}} * gamma + beta
+    This module differs from the built-in PyTorch BatchNorm3d as the mean and
+    standard-deviation are reduced across all devices during training.
+    For example, when one uses `nn.DataParallel` to wrap the network during
+    training, PyTorch's implementation normalize the tensor on each device using
+    the statistics only on that device, which accelerated the computation and
+    is also easy to implement, but the statistics might be inaccurate.
+    Instead, in this synchronized version, the statistics will be computed
+    over all training samples distributed on multiple devices.
+
+    Note that, for one-GPU or CPU-only case, this module behaves exactly same
+    as the built-in PyTorch implementation.
+    The mean and standard-deviation are calculated per-dimension over
+    the mini-batches and gamma and beta are learnable parameter vectors
+    of size C (where C is the input size).
+    During training, this layer keeps a running estimate of its computed mean
+    and variance. The running sum is kept with a default momentum of 0.1.
+    During evaluation, this running mean/variance is used for normalization.
+    Because the BatchNorm is done over the `C` dimension, computing statistics
+    on `(N, D, H, W)` slices, it's common terminology to call this Volumetric BatchNorm
+    or Spatio-temporal BatchNorm
+    Args:
+        num_features: num_features from an expected input of
+            size batch_size x num_features x depth x height x width
+        eps: a value added to the denominator for numerical stability.
+            Default: 1e-5
+        momentum: the value used for the running_mean and running_var
+            computation. Default: 0.1
+        affine: a boolean value that when set to ``True``, gives the layer learnable
+            affine parameters. Default: ``True``
+    Shape:
+        - Input: :math:`(N, C, D, H, W)`
+        - Output: :math:`(N, C, D, H, W)` (same shape as input)
+    Examples:
+        >>> # With Learnable Parameters
+        >>> m = SynchronizedBatchNorm3d(100)
+        >>> # Without Learnable Parameters
+        >>> m = SynchronizedBatchNorm3d(100, affine=False)
+        >>> input = torch.autograd.Variable(torch.randn(20, 100, 35, 45, 10))
+        >>> output = m(input)
+    """
+
+    def _check_input_dim(self, input):
+        if input.dim() != 5:
+            raise ValueError('expected 5D input (got {}D input)'
+                             .format(input.dim()))
+        super(SynchronizedBatchNorm3d, self)._check_input_dim(input)
\ No newline at end of file
diff --git a/modeling/sync_batchnorm/comm.py b/modeling/sync_batchnorm/comm.py
new file mode 100644
index 0000000..6005430
--- /dev/null
+++ b/modeling/sync_batchnorm/comm.py
@@ -0,0 +1,133 @@
+# -*- coding: utf-8 -*-
+# File   : comm.py
+# Author : Jiayuan Mao
+# Email  : maojiayuan@gmail.com
+# Date   : 27/01/2018
+#
+# This file is part of Synchronized-BatchNorm-PyTorch.
+# https://github.com/vacancy/Synchronized-BatchNorm-PyTorch
+# Distributed under MIT License.
+
+import sys
+if sys.version > '3':
+    import queue
+else:
+    import Queue as queue
+import collections
+import threading
+
+__all__ = ['FutureResult', 'SlavePipe', 'SyncMaster']
+
+
+class FutureResult(object):
+    """A thread-safe future implementation. Used only as one-to-one pipe."""
+
+    def __init__(self):
+        self._result = None
+        self._lock = threading.Lock()
+        self._cond = threading.Condition(self._lock)
+
+    def put(self, result):
+        with self._lock:
+            assert self._result is None, 'Previous result has\'t been fetched.'
+            self._result = result
+            self._cond.notify()
+
+    def get(self):
+        with self._lock:
+            if self._result is None:
+                self._cond.wait()
+
+            res = self._result
+            self._result = None
+            return res
+
+
+_MasterRegistry = collections.namedtuple('MasterRegistry', ['result'])
+_SlavePipeBase = collections.namedtuple('_SlavePipeBase', ['identifier', 'queue', 'result'])
+
+
+class SlavePipe(_SlavePipeBase):
+    """Pipe for master-slave communication."""
+
+    def run_slave(self, msg):
+        self.queue.put((self.identifier, msg))
+        ret = self.result.get()
+        self.queue.put(True)
+        return ret
+
+
+class SyncMaster(object):
+    """An abstract `SyncMaster` object.
+    - During the replication, as the data parallel will trigger an callback of each module, all slave devices should
+    call `register(id)` and obtain an `SlavePipe` to communicate with the master.
+    - During the forward pass, master device invokes `run_master`, all messages from slave devices will be collected,
+    and passed to a registered callback.
+    - After receiving the messages, the master device should gather the information and determine to message passed
+    back to each slave devices.
+    """
+
+    def __init__(self, master_callback):
+        """
+        Args:
+            master_callback: a callback to be invoked after having collected messages from slave devices.
+        """
+        self._master_callback = master_callback
+        self._queue = queue.Queue()
+        self._registry = collections.OrderedDict()
+        self._activated = False
+
+    def __getstate__(self):
+        return {'master_callback': self._master_callback}
+
+    def __setstate__(self, state):
+        self.__init__(state['master_callback'])
+
+    def register_slave(self, identifier):
+        """
+        Register an slave device.
+        Args:
+            identifier: an identifier, usually is the device id.
+        Returns: a `SlavePipe` object which can be used to communicate with the master device.
+        """
+        if self._activated:
+            assert self._queue.empty(), 'Queue is not clean before next initialization.'
+            self._activated = False
+            self._registry.clear()
+        future = FutureResult()
+        self._registry[identifier] = _MasterRegistry(future)
+        return SlavePipe(identifier, self._queue, future)
+
+    def run_master(self, master_msg):
+        """
+        Main entry for the master device in each forward pass.
+        The messages were first collected from each devices (including the master device), and then
+        an callback will be invoked to compute the message to be sent back to each devices
+        (including the master device).
+        Args:
+            master_msg: the message that the master want to send to itself. This will be placed as the first
+            message when calling `master_callback`. For detailed usage, see `_SynchronizedBatchNorm` for an example.
+        Returns: the message to be sent back to the master device.
+        """
+        self._activated = True
+
+        intermediates = [(0, master_msg)]
+        for i in range(self.nr_slaves):
+            intermediates.append(self._queue.get())
+
+        results = self._master_callback(intermediates)
+        assert results[0][0] == 0, 'The first result should belongs to the master.'
+
+        for i, res in results:
+            if i == 0:
+                continue
+            self._registry[i].result.put(res)
+
+        for i in range(self.nr_slaves):
+            assert self._queue.get() is True
+
+        return results[0][1]
+
+    @property
+    def nr_slaves(self):
+        return len(self._registry)
diff --git a/modeling/sync_batchnorm/replicate.py b/modeling/sync_batchnorm/replicate.py
new file mode 100644
index 0000000..3734266
--- /dev/null
+++ b/modeling/sync_batchnorm/replicate.py
@@ -0,0 +1,88 @@
+# -*- coding: utf-8 -*-
+# File   : replicate.py
+# Author : Jiayuan Mao
+# Email  : maojiayuan@gmail.com
+# Date   : 27/01/2018
+#
+# This file is part of Synchronized-BatchNorm-PyTorch.
+# https://github.com/vacancy/Synchronized-BatchNorm-PyTorch
+# Distributed under MIT License.
+
+import functools
+
+from torch.nn.parallel.data_parallel import DataParallel
+
+__all__ = [
+    'CallbackContext',
+    'execute_replication_callbacks',
+    'DataParallelWithCallback',
+    'patch_replication_callback'
+]
+
+
+class CallbackContext(object):
+    pass
+
+
+def execute_replication_callbacks(modules):
+    """
+    Execute an replication callback `__data_parallel_replicate__` on each module created by original replication.
+    The callback will be invoked with arguments `__data_parallel_replicate__(ctx, copy_id)`
+    Note that, as all modules are isomorphism, we assign each sub-module with a context
+    (shared among multiple copies of this module on different devices).
+    Through this context, different copies can share some information.
+    We guarantee that the callback on the master copy (the first copy) will be called ahead of calling the callback
+    of any slave copies.
+    """
+    master_copy = modules[0]
+    nr_modules = len(list(master_copy.modules()))
+    ctxs = [CallbackContext() for _ in range(nr_modules)]
+
+    for i, module in enumerate(modules):
+        for j, m in enumerate(module.modules()):
+            if hasattr(m, '__data_parallel_replicate__'):
+                m.__data_parallel_replicate__(ctxs[j], i)
+
+
+class DataParallelWithCallback(DataParallel):
+    """
+    Data Parallel with a replication callback.
+    An replication callback `__data_parallel_replicate__` of each module will be invoked after being created by
+    original `replicate` function.
+    The callback will be invoked with arguments `__data_parallel_replicate__(ctx, copy_id)`
+    Examples:
+        > sync_bn = SynchronizedBatchNorm1d(10, eps=1e-5, affine=False)
+        > sync_bn = DataParallelWithCallback(sync_bn, device_ids=[0, 1])
+        # sync_bn.__data_parallel_replicate__ will be invoked.
+    """
+
+    def replicate(self, module, device_ids):
+        modules = super(DataParallelWithCallback, self).replicate(module, device_ids)
+        execute_replication_callbacks(modules)
+        return modules
+
+
+def patch_replication_callback(data_parallel):
+    """
+    Monkey-patch an existing `DataParallel` object. Add the replication callback.
+    Useful when you have customized `DataParallel` implementation.
+    Examples:
+        > sync_bn = SynchronizedBatchNorm1d(10, eps=1e-5, affine=False)
+        > sync_bn = DataParallel(sync_bn, device_ids=[0, 1])
+        > patch_replication_callback(sync_bn)
+        # this is equivalent to
+        > sync_bn = SynchronizedBatchNorm1d(10, eps=1e-5, affine=False)
+        > sync_bn = DataParallelWithCallback(sync_bn, device_ids=[0, 1])
+    """
+
+    assert isinstance(data_parallel, DataParallel)
+
+    old_replicate = data_parallel.replicate
+
+    @functools.wraps(old_replicate)
+    def new_replicate(module, device_ids):
+        modules = old_replicate(module, device_ids)
+        execute_replication_callbacks(modules)
+        return modules
+
+    data_parallel.replicate = new_replicate
\ No newline at end of file
diff --git a/modeling/sync_batchnorm/unittest.py b/modeling/sync_batchnorm/unittest.py
new file mode 100644
index 0000000..f826560
--- /dev/null
+++ b/modeling/sync_batchnorm/unittest.py
@@ -0,0 +1,29 @@
+# -*- coding: utf-8 -*-
+# File   : unittest.py
+# Author : Jiayuan Mao
+# Email  : maojiayuan@gmail.com
+# Date   : 27/01/2018
+#
+# This file is part of Synchronized-BatchNorm-PyTorch.
+# https://github.com/vacancy/Synchronized-BatchNorm-PyTorch
+# Distributed under MIT License.
+
+import unittest
+
+import numpy as np
+from torch.autograd import Variable
+
+
+def as_numpy(v):
+    if isinstance(v, Variable):
+        v = v.data
+    return v.cpu().numpy()
+
+
+class TorchTestCase(unittest.TestCase):
+    def assertTensorClose(self, a, b, atol=1e-3, rtol=1e-3):
+        npa, npb = as_numpy(a), as_numpy(b)
+        self.assertTrue(
+                np.allclose(npa, npb, atol=atol),
+                'Tensor close check failed\n{}\n{}\nadiff={}, rdiff={}'.format(a, b, np.abs(npa - npb).max(), np.abs((npa - npb) / np.fmax(npa, 1e-5)).max())
+        )
diff --git a/train.py b/train.py
new file mode 100644
index 0000000..518e5df
--- /dev/null
+++ b/train.py
@@ -0,0 +1,48 @@
+from torchvision import transforms, utils
+from torch.utils.data import Dataset, DataLoader
+import matplotlib.pyplot as plt
+from scipy.io import loadmat
+from PIL import Image
+from deeplab import DeepLab
+
+
+DATADIR = "../data20200111/"
+
+class MyDataset(Dataset):
+    def __init__(self, imgtxt, labeltxt, transform=None, target_transform=None):
+        fh = open(imgtxt, 'r').readlines()
+        fhlabel = open(labeltxt, 'r').readlines()
+        imgs = []
+        for i in range(len(fh)):
+            print(i)
+            lineimg = fh[i].strip('\n')
+            linelabel = fhlabel[i].strip('\n')
+            imgs.append((lineimg, linelabel))
+        self.imgs = imgs
+        self.transform = transform
+        self.target_transform = target_transform
+        
+ 
+    def __getitem__(self, index):
+        fn, label = self.imgs[index]
+        img = loadmat(DATADIR+fn)['data']
+        label = loadmat(DATADIR+label)['data']
+        if self.transform is not None:
+            img = self.transform(img)
+            label = self.transform(label)
+        return img,label
+ 
+    def __len__(self):
+        return len(self.imgs)
+ 
+train_data=MyDataset(imgtxt='../data20200111/img_list.txt', labeltxt='../data20200111/label_list.txt', transform=transforms.ToTensor())
+data_loader = DataLoader(train_data, batch_size=16,shuffle=True)
+for epoch in range(10):
+    print("epoch, ", epoch)
+    for i, data in enumerate(data_loader):
+        inputs, labels = data
+        model = DeepLab(backbone='mobilenet', output_stride=16)
+        model.eval()
+        print(inputs.shape)
+        output = model(inputs)
+        print(output.size)
diff --git a/visualMat.py b/visualMat.py
new file mode 100644
index 0000000..55f09f6
--- /dev/null
+++ b/visualMat.py
@@ -0,0 +1,32 @@
+import os 
+import matplotlib.pyplot as plt
+from scipy.io import loadmat
+import argparse
+
+def readMatListandSave(path, name):
+    for obj in open(path+name).readlines():
+        obj = obj.strip("\n")
+        mat = loadmat(path+obj)['data']
+        plt.imshow(mat)
+        plt.savefig(path+obj.split('.')[0]+".png")
+        print("save ",path+obj.split('.')[0]+".png")
+    print("finish")
+
+def showPic(path, name):
+    mat = loadmat(path+name)['data']
+    plt.imshow(mat)
+    plt.show()
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description='input mat name')   # 首先创建一个ArgumentParser对象
+    parser.add_argument('path', type=str, help='please input mat or txt path')
+    parser.add_argument('name', type=str, help='please input file name')
+
+    args = parser.parse_args()
+    path = args.path
+    name = args.name
+    if name.split('.')[-1] == "txt": #接受list文件可视化并保存
+        readMatListandSave(path, name)
+    else:
+        showPic(path, name)