model = nn.Sequential(
            collections.OrderedDict(
            [
                        ("conv1_1", nn.Conv2d(in_channels=3,
                                  out_channels=64,
                                  kernel_size=(3, 3),
                                  stride=(1, 1),
                                  # (1(32-1)- 32 + 3)/2 = 1
                                  padding=1)),
                        ("relu1_1", nn.ReLU()),
                        ("conv1_2", nn.Conv2d(in_channels=64,
                                  out_channels=64,
                                  kernel_size=(3, 3),
                                  stride=(1, 1),
                                  padding=1)),

                        ("relu1_2", nn.ReLU()),
                        ("pool1", nn.MaxPool2d(kernel_size=(2, 2),
                                     stride=(2, 2))),

                        ("conv2_1", nn.Conv2d(in_channels=64,
                                  out_channels=128,
                                  kernel_size=(3, 3),
                                  stride=(1, 1),
                                  padding=1)),
                        ("relu_2_1", nn.ReLU()),

                        ("conv2_2", nn.Conv2d(in_channels=128,
                                  out_channels=128,
                                  kernel_size=(3, 3),
                                  stride=(1, 1),
                                  padding=1)),

                        ("relu2_2", nn.ReLU()),
                        ("pool2", nn.MaxPool2d(kernel_size=(2, 2),
                                     stride=(2, 2))),
                        ("conv3_1", nn.Conv2d(in_channels=128,
                        out_channels=256,
                        kernel_size=(3, 3),
                        stride=(1, 1),
                        padding=1)),
                        ("relu3_1", nn.ReLU()),
                        ("conv3_2", nn.Conv2d(in_channels=256,
                                  out_channels=256,
                                  kernel_size=(3, 3),
                                  stride=(1, 1),
                                  padding=1)),
                        ("relu3_2", nn.ReLU()),
                        ("conv3_3", nn.Conv2d(in_channels=256,
                                  out_channels=256,
                                  kernel_size=(3, 3),
                                  stride=(1, 1),
                                  padding=1)),
                        ("relu3_3", nn.ReLU()),
                        ("pool3", nn.MaxPool2d(kernel_size=(2, 2),
                                     stride=(2, 2))),
                        ("conv4_1", nn.Conv2d(in_channels=256,
                              out_channels=512,
                              kernel_size=(3, 3),
                              stride=(1, 1),
                              padding=1)),
                        ("relu4_1", nn.ReLU()),
                        ("conv4_2", nn.Conv2d(in_channels=512,
                                  out_channels=512,
                                  kernel_size=(3, 3),
                                  stride=(1, 1),
                                  padding=1)),
                        ("relu4_2", nn.ReLU()),
                        ("conv4_3", nn.Conv2d(in_channels=512,
                                  out_channels=512,
                                  kernel_size=(3, 3),
                                  stride=(1, 1),
                                  padding=1)),
                        ("relu4_3", nn.ReLU()),
                        ("pool4", nn.MaxPool2d(kernel_size=(2, 2),
                                     stride=(2, 2))),
                        ("conv5_1", nn.Conv2d(in_channels=512,
                                  out_channels=512,
                                  kernel_size=(3, 3),
                                  stride=(1, 1),
                                  padding=1)),
                        ("relu5_1", nn.ReLU()),
                        ("conv5_2", nn.Conv2d(in_channels=512,
                                  out_channels=512,
                                  kernel_size=(3, 3),
                        # Written by Ross Girshick and Sean Bell
          stride=(1, 1))),
                        ("relu5_2", nn.ReLU()),
                        ("conv5_3", nn.Conv2d(in_channels=512,
                                  out_channels=512,
                                  kernel_size=(3, 3),
                                  stride=(1, 1),
                                  padding=1)),
                        ("relu5_3", nn.ReLU()),
                        ("pool5", nn.MaxPool2d(kernel_size=(2, 2),
                                     stride=(2, 2))),
                        ("rpn_conv/3x3", nn.Conv2d(in_channels=512, out_channels=512, kernel_size=3, padding=1, stride=1 )),
                        ("rpn_relu/3x3", nn.ReLU()),
                        ("rpn_cls_score", nn.Conv2d(in_channels=512, out_channels=18, kernel_size=1,padding=0, stride=1)),
                        ("rpn_bbox_pred", nn.Conv2d(in_channels =512, out_channels=36, kernel_size=1, padding=0, stride=1)),
                        ("rpn_cls_score_reshape", Reshape([0, 2, -1, 0])),
                        ("rpn_cls_prob", nn.Softmax()),
                        ("rpn_cls_rpob_reshape", Reshape([0, 18, -1, 0])),
                        ("proposal", ProposalLayer(feat_stride=16, scales=1737, ratios=[0.5,1,2])),
                        ("roi_pool5", RoIPool((7,7),  spatial_scale=0.0625)),
                        ("fc6", nn.Linear(25088, 4096)),
                        ("relu6", nn.ReLU()),
                        ("drop6", nn.Dropout(0.5)),
                        ("fc7", nn.Linear(4096, 4096)),
                        ("drop7", nn.Dropout(0.5)),
                        ("cls_score", nn.Linear(4096, 3)),
                        ("bbox_pred", nn.Linear(4096, 12)),
                        ("cls_prob", nn.Softmax())
                ]
            )
)