Flexible DenseNet for pytorch

densenet.py

import re
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.utils.checkpoint as cp
from collections import OrderedDict
from torchvision.models.utils import load_state_dict_from_url


__all__ = ['DenseNet', 'densenet121', 'densenet169', 'densenet201', 'densenet161']

model_urls = {
    'densenet121': 'https://download.pytorch.org/models/densenet121-a639ec97.pth',
    'densenet169': 'https://download.pytorch.org/models/densenet169-b2777c0a.pth',
    'densenet201': 'https://download.pytorch.org/models/densenet201-c1103571.pth',
    'densenet161': 'https://download.pytorch.org/models/densenet161-8d451a50.pth',
}


def _bn_function_factory(norm, relu, conv):
    def bn_function(*inputs):
        concated_features = torch.cat(inputs, 1)
        bottleneck_output = conv(relu(norm(concated_features)))
        return bottleneck_output

    return bn_function


class _DenseLayer(nn.Sequential):
    def __init__(self, num_input_features, growth_rate, bn_size, drop_rate, memory_efficient=False, three_d=False):
        super(_DenseLayer, self).__init__()

        conv_layer = nn.Conv3d if three_d else nn.Conv2d
        batchnorm_layer = nn.BatchNorm3d if three_d else  nn.BatchNorm2d
        self.dropout_layer = F.dropout3d if three_d else F.dropout2d

        self.add_module('norm1', batchnorm_layer(num_input_features)),
        self.add_module('relu1', nn.ReLU(inplace=True)),
        self.add_module('conv1', conv_layer(num_input_features, bn_size *
                                           growth_rate, kernel_size=1, stride=1,
                                           bias=False)),
        self.add_module('norm2', batchnorm_layer(bn_size * growth_rate)),
        self.add_module('relu2', nn.ReLU(inplace=True)),
        self.add_module('conv2', conv_layer(bn_size * growth_rate, growth_rate,
                                           kernel_size=3, stride=1, padding=1,
                                           bias=False)),
        self.drop_rate = drop_rate
        self.memory_efficient = memory_efficient

    def forward(self, *prev_features):
        bn_function = _bn_function_factory(self.norm1, self.relu1, self.conv1)
        if self.memory_efficient and any(prev_feature.requires_grad for prev_feature in prev_features):
            bottleneck_output = cp.checkpoint(bn_function, *prev_features)
        else:
            bottleneck_output = bn_function(*prev_features)
        new_features = self.conv2(self.relu2(self.norm2(bottleneck_output)))
        if self.drop_rate > 0:
            new_features = self.dropout_layer(new_features, p=self.drop_rate,
                                     training=self.training)
        return new_features


class _DenseBlock(nn.Module):
    def __init__(self, num_layers, num_input_features, bn_size, growth_rate, drop_rate, memory_efficient=False, three_d=False):
        super(_DenseBlock, self).__init__()
        for i in range(num_layers):
            layer = _DenseLayer(
                num_input_features + i * growth_rate,
                growth_rate=growth_rate,
                bn_size=bn_size,
                drop_rate=drop_rate,
                memory_efficient=memory_efficient,
                three_d=three_d
            )
            self.add_module('denselayer%d' % (i + 1), layer)

    def forward(self, init_features):
        features = [init_features]
        for name, layer in self.named_children():
            new_features = layer(*features)
            features.append(new_features)
        return torch.cat(features, 1)


class _Transition(nn.Sequential):
    def __init__(self, num_input_features, num_output_features, fdc=False, max_pool=False, three_d=False):
        super(_Transition, self).__init__()
        if not fdc:
            conv_layer = nn.Conv3d if three_d else nn.Conv2d
            batchnorm_layer = nn.BatchNorm3d if three_d else  nn.BatchNorm2d
            self.add_module('norm', batchnorm_layer(num_input_features))
            self.add_module('relu', nn.ReLU(inplace=True))
        
            self.add_module('conv', conv_layer(num_input_features, num_output_features,
                                              kernel_size=1, stride=1, bias=False))
        if max_pool:
            maxpool_layer = nn.MaxPool3d if three_d else nn.MaxPool2d
            self.add_module('pool', maxpool_layer(kernel_size=2, stride=2))
        else:
            avgpool_layer = nn.AvgPool3d if three_d else nn.AvgPool2d
            self.add_module('pool', avgpool_layer(kernel_size=2, stride=2))


class DenseNet(nn.Module):
    r"""Densenet-BC model class, based on
    `"Densely Connected Convolutional Networks" <https://arxiv.org/pdf/1608.06993.pdf>`_
    Args:
        growth_rate (int) - how many filters to add each layer (`k` in paper)
        block_config (list of 4 ints) - how many layers in each pooling block
        input_channels (int) - 1 for grayscale, 3 for RGB (default: 3)
        num_init_features (int) - the number of filters to learn in the first convolution layer
        init_conv_kernel (int) - kernel size of first conv (default: 7)
        init_conv_stride (int) - stride of first conv (default: 2)
        init_pool (bool) - pool or not after first conv (default: True)
        max_pool (bool) - use max pooling in transition layers (default: False - avg pooling)
        bn_size (int) - multiplicative factor for number of bottle neck layers
          (i.e. bn_size * k features in the bottleneck layer)
        drop_rate (float) - dropout rate after each dense layer
        num_classes (int) - number of classification classes
        memory_efficient (bool) - If True, uses checkpointing. Much more memory efficient,
          but slower. Default: *False*. See `"paper" <https://arxiv.org/pdf/1707.06990.pdf>`_
        fdc (bool) - fully dense connectivity (no 1x1 conv in transition layer)
        exp_grow (bool) - doubles growth rate as block grows (ex: grow_rate=32 for block 0, 64 for block 1, 128 for block 2...)
        three_d (bool) - uses 3D conv, assumes (N,C,D,H,W) input
    """

    def __init__(self, growth_rate=32, block_config=(6, 12, 24, 16),
                 num_init_features=64, bn_size=4, drop_rate=0, num_classes=1000, memory_efficient=False, input_channels=3, init_conv_kernel=7, init_conv_stride=2, init_pool=True, max_pool=False, fdc=False, exp_grow=False, three_d=False):

        super(DenseNet, self).__init__()

        conv_layer = nn.Conv3d if three_d else nn.Conv2d
        maxpool_layer = nn.MaxPool3d if three_d else nn.MaxPool2d
        batchnorm_layer = nn.BatchNorm3d if three_d else  nn.BatchNorm2d
        self.globalpool_layer = F.adaptive_avg_pool3d if three_d else  F.adaptive_avg_pool2d

        # First convolution
        self.features = nn.Sequential(OrderedDict([
            ('conv0', conv_layer(input_channels, num_init_features, kernel_size=init_conv_kernel, stride=init_conv_stride,
                                padding=init_conv_kernel//2, bias=False)),
            ('norm0', batchnorm_layer(num_init_features)),
            ('relu0', nn.ReLU(inplace=True)),
        ]))
        if init_pool:
            self.features.add_module('pool0', maxpool_layer(kernel_size=3, stride=2, padding=1))
        
        # Each denseblock
        num_features = num_init_features
        for i, num_layers in enumerate(block_config):
            block = _DenseBlock(
                num_layers=num_layers,
                num_input_features=num_features,
                bn_size=bn_size,
                growth_rate=(growth_rate * (2**i)) if exp_grow else growth_rate,
                drop_rate=drop_rate,
                memory_efficient=memory_efficient,
                three_d=three_d
            )
            self.features.add_module('denseblock%d' % (i + 1), block)
            num_features = (num_features + num_layers * growth_rate * (2**i)) if exp_grow else (num_features + num_layers * growth_rate)
            if i != len(block_config) - 1:
                trans = _Transition(num_input_features=num_features,
                                    num_output_features=num_features // 2,
                                    fdc=fdc,
                                    max_pool=max_pool)
                self.features.add_module('transition%d' % (i + 1), trans)
                if not fdc:
                    num_features = num_features // 2

        # Final batch norm
        self.features.add_module('norm5', batchnorm_layer(num_features))

        # Linear layer
        self.classifier = conv_layer(num_features, num_classes, 1)

        # Official init from torch repo.
        for m in self.modules():
            if isinstance(m, nn.Conv2d) or isinstance(m, nn.Conv3d):
                nn.init.kaiming_normal_(m.weight)
            elif isinstance(m, nn.BatchNorm2d) or isinstance(m, nn.BatchNorm3d):
                nn.init.constant_(m.weight, 1)
                nn.init.constant_(m.bias, 0)
            elif isinstance(m, nn.Linear):
                nn.init.constant_(m.bias, 0)

    def forward(self, x, heatmap=False):
        features = self.features(x)
        if not heatmap:
            out = F.relu(features, inplace=True)
            out = self.globalpool_layer(out, 1)
            out = self.classifier(out).flatten(1)
            return out
        else:
            out = F.relu(features, inplace=True)
            out = self.classifier(out)
            return self.globalpool_layer(out, 1).flatten(1), out


def _load_state_dict(model, model_url, progress):
    # '.'s are no longer allowed in module names, but previous _DenseLayer
    # has keys 'norm.1', 'relu.1', 'conv.1', 'norm.2', 'relu.2', 'conv.2'.
    # They are also in the checkpoints in model_urls. This pattern is used
    # to find such keys.
    pattern = re.compile(
        r'^(.*denselayer\d+\.(?:norm|relu|conv))\.((?:[12])\.(?:weight|bias|running_mean|running_var))$')

    state_dict = load_state_dict_from_url(model_url, progress=progress)
    for key in list(state_dict.keys()):
        res = pattern.match(key)
        if res:
            new_key = res.group(1) + res.group(2)
            state_dict[new_key] = state_dict[key]
            del state_dict[key]
    state_dict['classifier.weight'].unsqueeze_(-1).unsqueeze_(-1)
    model.load_state_dict(state_dict)


def _densenet(arch, growth_rate, block_config, num_init_features, pretrained, progress,
              **kwargs):
    model = DenseNet(growth_rate, block_config, num_init_features, **kwargs)
    if pretrained:
        _load_state_dict(model, model_urls[arch], progress)
    return model


def densenet121(pretrained=False, progress=True, **kwargs):
    r"""Densenet-121 model from
    `"Densely Connected Convolutional Networks" <https://arxiv.org/pdf/1608.06993.pdf>`_
    Args:
        pretrained (bool): If True, returns a model pre-trained on ImageNet
        progress (bool): If True, displays a progress bar of the download to stderr
        memory_efficient (bool) - If True, uses checkpointing. Much more memory efficient,
          but slower. Default: *False*. See `"paper" <https://arxiv.org/pdf/1707.06990.pdf>`_
    """
    return _densenet('densenet121', 32, (6, 12, 24, 16), 64, pretrained, progress,
                     **kwargs)


def densenet161(pretrained=False, progress=True, **kwargs):
    r"""Densenet-161 model from
    `"Densely Connected Convolutional Networks" <https://arxiv.org/pdf/1608.06993.pdf>`_
    Args:
        pretrained (bool): If True, returns a model pre-trained on ImageNet
        progress (bool): If True, displays a progress bar of the download to stderr
        memory_efficient (bool) - If True, uses checkpointing. Much more memory efficient,
          but slower. Default: *False*. See `"paper" <https://arxiv.org/pdf/1707.06990.pdf>`_
    """
    return _densenet('densenet161', 48, (6, 12, 36, 24), 96, pretrained, progress,
                     **kwargs)


def densenet169(pretrained=False, progress=True, **kwargs):
    r"""Densenet-169 model from
    `"Densely Connected Convolutional Networks" <https://arxiv.org/pdf/1608.06993.pdf>`_
    Args:
        pretrained (bool): If True, returns a model pre-trained on ImageNet
        progress (bool): If True, displays a progress bar of the download to stderr
        memory_efficient (bool) - If True, uses checkpointing. Much more memory efficient,
          but slower. Default: *False*. See `"paper" <https://arxiv.org/pdf/1707.06990.pdf>`_
    """
    return _densenet('densenet169', 32, (6, 12, 32, 32), 64, pretrained, progress,
                     **kwargs)


def densenet201(pretrained=False, progress=True, **kwargs):
    r"""Densenet-201 model from
    `"Densely Connected Convolutional Networks" <https://arxiv.org/pdf/1608.06993.pdf>`_
    Args:
        pretrained (bool): If True, returns a model pre-trained on ImageNet
        progress (bool): If True, displays a progress bar of the download to stderr
        memory_efficient (bool) - If True, uses checkpointing. Much more memory efficient,
          but slower. Default: *False*. See `"paper" <https://arxiv.org/pdf/1707.06990.pdf>`_
    """
    return _densenet('densenet201', 32, (6, 12, 48, 32), 64, pretrained, progress,
                     **kwargs)