graft-pytorch 0.1.7__py3-none-any.whl

graft/models/resnet.py

@@ -0,0 +1,564 @@
+'''ResNet in PyTorch.
+Reference
+    Deep Residual Learning for Image Recognition
+    https://arxiv.org/abs/1512.03385
+'''
+from typing import Any, Callable, List, Optional, Type, Union, TypeVar, Dict
+import torch
+import torch.nn as nn
+from torch import Tensor
+
+
+__all__ = [
+    "ResNet",
+    "ResNet18",
+    "ResNet34", 
+    "ResNet50",
+    "ResNet101",
+    "ResNet152",
+    "ResNext50_32x4d",
+    "ResNext101_32x8d",
+    "ResNext101_64x4d",
+    "wide_resnet50_2",
+    "wide_resnet101_2",
+]
+
+
+V = TypeVar("V")
+def _ovewrite_value_param(param: str, actual: Optional[V], expected: V) -> V:
+    if actual is not None:
+        if actual != expected:
+            raise ValueError(f"The parameter '{param}' expected value {expected} but got {actual} instead.")
+    return expected
+
+
+def _ovewrite_named_param(kwargs: Dict[str, Any], param: str, new_value: V) -> None:
+    if param in kwargs:
+        if kwargs[param] != new_value:
+            raise ValueError(f"The parameter '{param}' expected value {new_value} but got {kwargs[param]} instead.")
+    else:
+        kwargs[param] = new_value
+
+
+def conv3x3(in_planes: int, out_planes: int, stride: int = 1, groups: int = 1, dilation: int = 1) -> nn.Conv2d:
+    """3x3 convolution with padding"""
+    return nn.Conv2d(
+        in_planes,
+        out_planes,
+        kernel_size=3,
+        stride=stride,
+        padding=dilation,
+        groups=groups,
+        bias=False,
+        dilation=dilation,
+    )
+
+
+def conv1x1(in_planes: int, out_planes: int, stride: int = 1) -> nn.Conv2d:
+    """1x1 convolution"""
+    return nn.Conv2d(in_planes, out_planes, kernel_size=1, stride=stride, bias=False)
+
+
+class BasicBlock(nn.Module):
+    expansion: int = 1
+
+    def __init__(
+        self,
+        inplanes: int,
+        planes: int,
+        stride: int = 1,
+        downsample: Optional[nn.Module] = None,
+        groups: int = 1,
+        base_width: int = 64,
+        dilation: int = 1,
+        norm_layer: Optional[Callable[..., nn.Module]] = None,
+    ) -> None:
+        super().__init__()
+        if norm_layer is None:
+            norm_layer = nn.BatchNorm2d
+        if groups != 1 or base_width != 64:
+            raise ValueError("BasicBlock only supports groups=1 and base_width=64")
+        if dilation > 1:
+            raise NotImplementedError("Dilation > 1 not supported in BasicBlock")
+        # Both self.conv1 and self.downsample layers downsample the input when stride != 1
+        self.conv1 = conv3x3(inplanes, planes, stride)
+        self.bn1 = norm_layer(planes)
+        self.relu = nn.ReLU(inplace=True)
+        self.conv2 = conv3x3(planes, planes)
+        self.bn2 = norm_layer(planes)
+        self.downsample = downsample
+        self.stride = stride
+
+    def forward(self, x: Tensor) -> Tensor:
+        identity = 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:
+            identity = self.downsample(x)
+
+        out += identity
+        out = self.relu(out)
+
+        return out
+
+
+class Bottleneck(nn.Module):
+    # Bottleneck in torchvision places the stride for downsampling at 3x3 convolution(self.conv2)
+    # while original implementation places the stride at the first 1x1 convolution(self.conv1)
+    # according to "Deep residual learning for image recognition"https://arxiv.org/abs/1512.03385.
+    # This variant is also known as ResNet V1.5 and improves accuracy according to
+    # https://ngc.nvidia.com/catalog/model-scripts/nvidia:resnet_50_v1_5_for_pytorch.
+
+    expansion: int = 4
+
+    def __init__(
+        self,
+        inplanes: int,
+        planes: int,
+        stride: int = 1,
+        downsample: Optional[nn.Module] = None,
+        groups: int = 1,
+        base_width: int = 64,
+        dilation: int = 1,
+        norm_layer: Optional[Callable[..., nn.Module]] = None,
+    ) -> None:
+        super().__init__()
+        if norm_layer is None:
+            norm_layer = nn.BatchNorm2d
+        width = int(planes * (base_width / 64.0)) * groups
+        # Both self.conv2 and self.downsample layers downsample the input when stride != 1
+        self.conv1 = conv1x1(inplanes, width)
+        self.bn1 = norm_layer(width)
+        self.conv2 = conv3x3(width, width, stride, groups, dilation)
+        self.bn2 = norm_layer(width)
+        self.conv3 = conv1x1(width, planes * self.expansion)
+        self.bn3 = norm_layer(planes * self.expansion)
+        self.relu = nn.ReLU(inplace=True)
+        self.downsample = downsample
+        self.stride = stride
+
+    def forward(self, x: Tensor) -> Tensor:
+        identity = 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:
+            identity = self.downsample(x)
+
+        out += identity
+        out = self.relu(out)
+
+        return out
+
+
+class ResNet(nn.Module):
+    def __init__(
+        self,
+        block: Type[Union[BasicBlock, Bottleneck]],
+        layers: List[int],
+        num_classes: int = 1000,
+        zero_init_residual: bool = False,
+        groups: int = 1,
+        width_per_group: int = 64,
+        replace_stride_with_dilation: Optional[List[bool]] = None,
+        norm_layer: Optional[Callable[..., nn.Module]] = None,
+    ) -> None:
+        super().__init__()
+        # _log_api_usage_once(self)
+        if norm_layer is None:
+            norm_layer = nn.BatchNorm2d
+        self._norm_layer = norm_layer
+
+        self.inplanes = 64
+        self.dilation = 1
+        if replace_stride_with_dilation is None:
+            # each element in the tuple indicates if we should replace
+            # the 2x2 stride with a dilated convolution instead
+            replace_stride_with_dilation = [False, False, False]
+        if len(replace_stride_with_dilation) != 3:
+            raise ValueError(
+                "replace_stride_with_dilation should be None "
+                f"or a 3-element tuple, got {replace_stride_with_dilation}"
+            )
+        self.groups = groups
+        self.base_width = width_per_group
+        self.conv1 = nn.Conv2d(3, self.inplanes, kernel_size=7, stride=2, padding=3, bias=False)
+        self.bn1 = norm_layer(self.inplanes)
+        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, dilate=replace_stride_with_dilation[0])
+        self.layer3 = self._make_layer(block, 256, layers[2], stride=2, dilate=replace_stride_with_dilation[1])
+        self.layer4 = self._make_layer(block, 512, layers[3], stride=2, dilate=replace_stride_with_dilation[2])
+        self.avgpool = nn.AdaptiveAvgPool2d((1, 1))
+        self.fc = nn.Linear(512 * block.expansion, num_classes)
+        self.embDim = 512 * block.expansion
+        for m in self.modules():
+            if isinstance(m, nn.Conv2d):
+                nn.init.kaiming_normal_(m.weight, mode="fan_out", nonlinearity="relu")
+            elif isinstance(m, (nn.BatchNorm2d, nn.GroupNorm)):
+                nn.init.constant_(m.weight, 1)
+                nn.init.constant_(m.bias, 0)
+
+        # Zero-initialize the last BN in each residual branch,
+        # so that the residual branch starts with zeros, and each residual block behaves like an identity.
+        # This improves the model by 0.2~0.3% according to https://arxiv.org/abs/1706.02677
+        if zero_init_residual:
+            for m in self.modules():
+                if isinstance(m, Bottleneck) and m.bn3.weight is not None:
+                    nn.init.constant_(m.bn3.weight, 0)  # type: ignore[arg-type]
+                elif isinstance(m, BasicBlock) and m.bn2.weight is not None:
+                    nn.init.constant_(m.bn2.weight, 0)  # type: ignore[arg-type]
+
+    def _make_layer(
+        self,
+        block: Type[Union[BasicBlock, Bottleneck]],
+        planes: int,
+        blocks: int,
+        stride: int = 1,
+        dilate: bool = False,
+    ) -> nn.Sequential:
+        norm_layer = self._norm_layer
+        downsample = None
+        previous_dilation = self.dilation
+        if dilate:
+            self.dilation *= stride
+            stride = 1
+        if stride != 1 or self.inplanes != planes * block.expansion:
+            downsample = nn.Sequential(
+                conv1x1(self.inplanes, planes * block.expansion, stride),
+                norm_layer(planes * block.expansion),
+            )
+
+        layers = []
+        layers.append(
+            block(
+                self.inplanes, planes, stride, downsample, self.groups, self.base_width, previous_dilation, norm_layer
+            )
+        )
+        self.inplanes = planes * block.expansion
+        for _ in range(1, blocks):
+            layers.append(
+                block(
+                    self.inplanes,
+                    planes,
+                    groups=self.groups,
+                    base_width=self.base_width,
+                    dilation=self.dilation,
+                    norm_layer=norm_layer,
+                )
+            )
+
+        return nn.Sequential(*layers)
+
+    def _forward_impl(self, x: Tensor, last=False, freeze=False) -> Tensor:
+        # See note [TorchScript super()]
+        if freeze:
+            with torch.no_grad():
+                self.eval()
+                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)
+                features = torch.flatten(x, 1)
+                self.train()
+        else:
+            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)
+            features = torch.flatten(x, 1)
+
+        out = self.fc(features)
+        if last:
+            return out, features
+        else:
+            return out
+
+    def forward(self, x: Tensor, last=False, freeze=False) -> Tensor:
+        return self._forward_impl(x, last, freeze)
+
+    def get_embedding_dim(self):
+        return self.embDim
+
+    def get_grads(self) -> torch.Tensor:
+        """
+        Returns all the gradients concatenated in a single tensor.
+        :return: gradients tensor (??)
+        """
+        grads = []
+        for pp in list(self.parameters()):
+            if pp.requires_grad: # only using the parameter that require the gradient
+                grads.append(pp.grad.view(-1))
+        return torch.cat(grads)
+
+
+def _resnet(
+    block: Type[Union[BasicBlock, Bottleneck]],
+    layers: List[int],
+    **kwargs: Any,
+) -> ResNet:
+    model = ResNet(block, layers, **kwargs)
+    return model
+
+
+
+def ResNet18(num_classes: int, **kwargs: Any) -> ResNet:
+    """ResNet-18 from `Deep Residual Learning for Image Recognition <https://arxiv.org/pdf/1512.03385.pdf>`__.
+    Args:
+        weights (:class:`~torchvision.models.ResNet18_Weights`, optional): The
+            pretrained weights to use. See
+            :class:`~torchvision.models.ResNet18_Weights` below for
+            more details, and possible values. By default, no pre-trained
+            weights are used.
+        progress (bool, optional): If True, displays a progress bar of the
+            download to stderr. Default is True.
+        **kwargs: parameters passed to the ``torchvision.models.resnet.ResNet``
+            base class. Please refer to the `source code
+            <https://github.com/pytorch/vision/blob/main/torchvision/models/resnet.py>`_
+            for more details about this class.
+    .. autoclass:: torchvision.models.ResNet18_Weights
+        :members:
+    """
+    return _resnet(BasicBlock, [2, 2, 2, 2], num_classes=num_classes, **kwargs)
+
+def ResNet34(num_classes: int, **kwargs: Any) -> ResNet:
+    """ResNet-34 from `Deep Residual Learning for Image Recognition <https://arxiv.org/pdf/1512.03385.pdf>`__.
+    Args:
+        weights (:class:`~torchvision.models.ResNet34_Weights`, optional): The
+            pretrained weights to use. See
+            :class:`~torchvision.models.ResNet34_Weights` below for
+            more details, and possible values. By default, no pre-trained
+            weights are used.
+        progress (bool, optional): If True, displays a progress bar of the
+            download to stderr. Default is True.
+        **kwargs: parameters passed to the ``torchvision.models.resnet.ResNet``
+            base class. Please refer to the `source code
+            <https://github.com/pytorch/vision/blob/main/torchvision/models/resnet.py>`_
+            for more details about this class.
+    .. autoclass:: torchvision.models.ResNet34_Weights
+        :members:
+    """
+    return _resnet(BasicBlock, [3, 4, 6, 3], num_classes=num_classes, **kwargs)
+
+
+def ResNet50(num_classes: int, **kwargs: Any) -> ResNet:
+    """ResNet-50 from `Deep Residual Learning for Image Recognition <https://arxiv.org/pdf/1512.03385.pdf>`__.
+    .. note::
+       The bottleneck of TorchVision places the stride for downsampling to the second 3x3
+       convolution while the original paper places it to the first 1x1 convolution.
+       This variant improves the accuracy and is known as `ResNet V1.5
+       <https://ngc.nvidia.com/catalog/model-scripts/nvidia:resnet_50_v1_5_for_pytorch>`_.
+    Args:
+        weights (:class:`~torchvision.models.ResNet50_Weights`, optional): The
+            pretrained weights to use. See
+            :class:`~torchvision.models.ResNet50_Weights` below for
+            more details, and possible values. By default, no pre-trained
+            weights are used.
+        progress (bool, optional): If True, displays a progress bar of the
+            download to stderr. Default is True.
+        **kwargs: parameters passed to the ``torchvision.models.resnet.ResNet``
+            base class. Please refer to the `source code
+            <https://github.com/pytorch/vision/blob/main/torchvision/models/resnet.py>`_
+            for more details about this class.
+    .. autoclass:: torchvision.models.ResNet50_Weights
+        :members:
+    """
+    return _resnet(Bottleneck, [3, 4, 6, 3], num_classes=num_classes, **kwargs)
+
+
+
+def ResNet101(num_classes: int, **kwargs: Any) -> ResNet:
+    """ResNet-101 from `Deep Residual Learning for Image Recognition <https://arxiv.org/pdf/1512.03385.pdf>`__.
+    .. note::
+       The bottleneck of TorchVision places the stride for downsampling to the second 3x3
+       convolution while the original paper places it to the first 1x1 convolution.
+       This variant improves the accuracy and is known as `ResNet V1.5
+       <https://ngc.nvidia.com/catalog/model-scripts/nvidia:resnet_50_v1_5_for_pytorch>`_.
+    Args:
+        weights (:class:`~torchvision.models.ResNet101_Weights`, optional): The
+            pretrained weights to use. See
+            :class:`~torchvision.models.ResNet101_Weights` below for
+            more details, and possible values. By default, no pre-trained
+            weights are used.
+        progress (bool, optional): If True, displays a progress bar of the
+            download to stderr. Default is True.
+        **kwargs: parameters passed to the ``torchvision.models.resnet.ResNet``
+            base class. Please refer to the `source code
+            <https://github.com/pytorch/vision/blob/main/torchvision/models/resnet.py>`_
+            for more details about this class.
+    .. autoclass:: torchvision.models.ResNet101_Weights
+        :members:
+    """
+    return _resnet(Bottleneck, [3, 4, 23, 3], num_classes=num_classes, **kwargs)
+
+
+def ResNet152(num_classes: int, **kwargs: Any) -> ResNet:
+    """ResNet-152 from `Deep Residual Learning for Image Recognition <https://arxiv.org/pdf/1512.03385.pdf>`__.
+    .. note::
+       The bottleneck of TorchVision places the stride for downsampling to the second 3x3
+       convolution while the original paper places it to the first 1x1 convolution.
+       This variant improves the accuracy and is known as `ResNet V1.5
+       <https://ngc.nvidia.com/catalog/model-scripts/nvidia:resnet_50_v1_5_for_pytorch>`_.
+    Args:
+        weights (:class:`~torchvision.models.ResNet152_Weights`, optional): The
+            pretrained weights to use. See
+            :class:`~torchvision.models.ResNet152_Weights` below for
+            more details, and possible values. By default, no pre-trained
+            weights are used.
+        progress (bool, optional): If True, displays a progress bar of the
+            download to stderr. Default is True.
+        **kwargs: parameters passed to the ``torchvision.models.resnet.ResNet``
+            base class. Please refer to the `source code
+            <https://github.com/pytorch/vision/blob/main/torchvision/models/resnet.py>`_
+            for more details about this class.
+    .. autoclass:: torchvision.models.ResNet152_Weights
+        :members:
+    """
+    return _resnet(Bottleneck, [3, 8, 36, 3], num_classes=num_classes, **kwargs)
+
+
+def ResNext50_32x4d(num_classes: int, **kwargs: Any) -> ResNet:
+    """ResNeXt-50 32x4d model from
+    `Aggregated Residual Transformation for Deep Neural Networks <https://arxiv.org/abs/1611.05431>`_.
+    Args:
+        weights (:class:`~torchvision.models.ResNeXt50_32X4D_Weights`, optional): The
+            pretrained weights to use. See
+            :class:`~torchvision.models.ResNext50_32X4D_Weights` below for
+            more details, and possible values. By default, no pre-trained
+            weights are used.
+        progress (bool, optional): If True, displays a progress bar of the
+            download to stderr. Default is True.
+        **kwargs: parameters passed to the ``torchvision.models.resnet.ResNet``
+            base class. Please refer to the `source code
+            <https://github.com/pytorch/vision/blob/main/torchvision/models/resnet.py>`_
+            for more details about this class.
+    .. autoclass:: torchvision.models.ResNeXt50_32X4D_Weights
+        :members:
+    """
+    _ovewrite_named_param(kwargs, "groups", 32)
+    _ovewrite_named_param(kwargs, "width_per_group", 4)
+    return _resnet(Bottleneck, [3, 4, 6, 3], num_classes=num_classes, **kwargs)
+
+
+def ResNext101_32x8d(num_classes: int, **kwargs: Any) -> ResNet:
+    """ResNeXt-101 32x8d model from
+    `Aggregated Residual Transformation for Deep Neural Networks <https://arxiv.org/abs/1611.05431>`_.
+    Args:
+        weights (:class:`~torchvision.models.ResNeXt101_32X8D_Weights`, optional): The
+            pretrained weights to use. See
+            :class:`~torchvision.models.ResNeXt101_32X8D_Weights` below for
+            more details, and possible values. By default, no pre-trained
+            weights are used.
+        progress (bool, optional): If True, displays a progress bar of the
+            download to stderr. Default is True.
+        **kwargs: parameters passed to the ``torchvision.models.resnet.ResNet``
+            base class. Please refer to the `source code
+            <https://github.com/pytorch/vision/blob/main/torchvision/models/resnet.py>`_
+            for more details about this class.
+    .. autoclass:: torchvision.models.ResNeXt101_32X8D_Weights
+        :members:
+    """
+    _ovewrite_named_param(kwargs, "groups", 32)
+    _ovewrite_named_param(kwargs, "width_per_group", 8)
+    return _resnet(Bottleneck, [3, 4, 23, 3], num_classes=num_classes, **kwargs)
+
+
+def ResNext101_64x4d(num_classes: int, **kwargs: Any) -> ResNet:
+    """ResNeXt-101 64x4d model from
+    `Aggregated Residual Transformation for Deep Neural Networks <https://arxiv.org/abs/1611.05431>`_.
+    Args:
+        weights (:class:`~torchvision.models.ResNeXt101_64X4D_Weights`, optional): The
+            pretrained weights to use. See
+            :class:`~torchvision.models.ResNeXt101_64X4D_Weights` below for
+            more details, and possible values. By default, no pre-trained
+            weights are used.
+        progress (bool, optional): If True, displays a progress bar of the
+            download to stderr. Default is True.
+        **kwargs: parameters passed to the ``torchvision.models.resnet.ResNet``
+            base class. Please refer to the `source code
+            <https://github.com/pytorch/vision/blob/main/torchvision/models/resnet.py>`_
+            for more details about this class.
+    .. autoclass:: torchvision.models.ResNeXt101_64X4D_Weights
+        :members:
+    """
+    _ovewrite_named_param(kwargs, "groups", 64)
+    _ovewrite_named_param(kwargs, "width_per_group", 4)
+    return _resnet(Bottleneck, [3, 4, 23, 3], num_classes=num_classes, **kwargs)
+
+
+def wide_resnet50_2(num_classes: int, **kwargs: Any) -> ResNet:
+    """Wide ResNet-50-2 model from
+    `Wide Residual Networks <https://arxiv.org/abs/1605.07146>`_.
+    The model is the same as ResNet except for the bottleneck number of channels
+    which is twice larger in every block. The number of channels in outer 1x1
+    convolutions is the same, e.g. last block in ResNet-50 has 2048-512-2048
+    channels, and in Wide ResNet-50-2 has 2048-1024-2048.
+    Args:
+        weights (:class:`~torchvision.models.Wide_ResNet50_2_Weights`, optional): The
+            pretrained weights to use. See
+            :class:`~torchvision.models.Wide_ResNet50_2_Weights` below for
+            more details, and possible values. By default, no pre-trained
+            weights are used.
+        progress (bool, optional): If True, displays a progress bar of the
+            download to stderr. Default is True.
+        **kwargs: parameters passed to the ``torchvision.models.resnet.ResNet``
+            base class. Please refer to the `source code
+            <https://github.com/pytorch/vision/blob/main/torchvision/models/resnet.py>`_
+            for more details about this class.
+    .. autoclass:: torchvision.models.Wide_ResNet50_2_Weights
+        :members:
+    """
+    _ovewrite_named_param(kwargs, "width_per_group", 64 * 2)
+    return _resnet(Bottleneck, [3, 4, 6, 3], num_classes=num_classes, **kwargs)
+
+
+def wide_resnet101_2(num_classes: int, **kwargs: Any) -> ResNet:
+    """Wide ResNet-101-2 model from
+    `Wide Residual Networks <https://arxiv.org/abs/1605.07146>`_.
+    The model is the same as ResNet except for the bottleneck number of channels
+    which is twice larger in every block. The number of channels in outer 1x1
+    convolutions is the same, e.g. last block in ResNet-101 has 2048-512-2048
+    channels, and in Wide ResNet-101-2 has 2048-1024-2048.
+    Args:
+        weights (:class:`~torchvision.models.Wide_ResNet101_2_Weights`, optional): The
+            pretrained weights to use. See
+            :class:`~torchvision.models.Wide_ResNet101_2_Weights` below for
+            more details, and possible values. By default, no pre-trained
+            weights are used.
+        progress (bool, optional): If True, displays a progress bar of the
+            download to stderr. Default is True.
+        **kwargs: parameters passed to the ``torchvision.models.resnet.ResNet``
+            base class. Please refer to the `source code
+            <https://github.com/pytorch/vision/blob/main/torchvision/models/resnet.py>`_
+            for more details about this class.
+    .. autoclass:: torchvision.models.Wide_ResNet101_2_Weights
+        :members:
+    """
+    _ovewrite_named_param(kwargs, "width_per_group", 64 * 2)
+    return _resnet(Bottleneck, [3, 4, 23, 3], num_classes=num_classes, **kwargs)

graft/models/resnet9.py

@@ -0,0 +1,72 @@
+import torch.nn as nn
+import torch
+from torch import Tensor
+
+
+def conv_block(in_channels, out_channels, pool=False):
+    layers = [nn.Conv2d(in_channels, out_channels, kernel_size=3, padding=1), 
+              nn.BatchNorm2d(out_channels), 
+              nn.ReLU(inplace=True)]
+    if pool: layers.append(nn.MaxPool2d(2))
+    return nn.Sequential(*layers)
+
+class ResNet9(nn.Module):
+    def __init__(self, in_channels, num_classes):
+        super().__init__()
+        
+        self.conv1 = conv_block(in_channels, 64)
+        self.conv2 = conv_block(64, 128, pool=True)
+        self.res1 = nn.Sequential(conv_block(128, 128), conv_block(128, 128))
+        
+        self.conv3 = conv_block(128, 256, pool=True)
+        self.conv4 = conv_block(256, 512, pool=True)
+        self.res2 = nn.Sequential(conv_block(512, 512), conv_block(512, 512))
+        
+        self.pool = nn.MaxPool2d(4)
+        self.fc = nn.Linear(512, num_classes)
+#         self.classifier = nn.Sequential(nn.MaxPool2d(4), 
+#                                         nn.Flatten(), 
+#                                         nn.Linear(512, num_classes))
+        
+#     def forward(self, xb):
+#         out = self.conv1(xb)
+#         out = self.conv2(out)
+#         out = self.res1(out) + out
+#         out = self.conv3(out)
+#         out = self.conv4(out)
+#         out = self.res2(out) + out
+#         out = self.classifier(out)
+#         return out
+    
+    def _forward_impl(self, x: Tensor, last=False, freeze=False) -> Tensor:
+        # See note [TorchScript super()]
+        if freeze:
+            with torch.no_grad():
+                self.eval()
+                x = self.conv1(x)
+                x = self.conv2(x)
+                x = self.res1(x) + x
+                x = self.conv3(x)
+                x = self.conv4(x)
+                x = self.res2(x) + x
+                x = self.pool(x)
+                features = torch.flatten(x, 1)
+                self.train()
+        else:
+            x = self.conv1(x)
+            x = self.conv2(x)
+            x = self.res1(x) + x
+            x = self.conv3(x)
+            x = self.conv4(x)
+            x = self.res2(x) + x
+            x = self.pool(x)
+            features = torch.flatten(x, 1)
+
+        out = self.fc(features)
+        if last:
+            return out, features
+        else:
+            return out
+
+    def forward(self, x: Tensor, last=False, freeze=False) -> Tensor:
+        return self._forward_impl(x, last, freeze)