densenet_features.py

import re
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.utils.model_zoo as model_zoo
from collections import OrderedDict


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',
}
model_dir = './pretrained_models'


class _DenseLayer(nn.Sequential):

    num_layers = 2

    def __init__(self, num_input_features, growth_rate, bn_size, drop_rate):
        super(_DenseLayer, self).__init__()
        self.add_module('norm1', nn.BatchNorm2d(num_input_features)),
        self.add_module('relu1', nn.ReLU(inplace=True)),
        self.add_module('conv1', nn.Conv2d(num_input_features, bn_size *
                        growth_rate, kernel_size=1, stride=1, bias=False)),
        self.add_module('norm2', nn.BatchNorm2d(bn_size * growth_rate)),
        self.add_module('relu2', nn.ReLU(inplace=True)),
        self.add_module('conv2', nn.Conv2d(bn_size * growth_rate, growth_rate,
                        kernel_size=3, stride=1, padding=1, bias=False)),
        self.drop_rate = drop_rate

    def forward(self, x):
        new_features = super(_DenseLayer, self).forward(x)
        if self.drop_rate > 0:
            new_features = F.dropout(new_features, p=self.drop_rate, training=self.training)

        # channelwise concatenation
        return torch.cat([x, new_features], 1)

    def layer_conv_info(self):
        layer_kernel_sizes = [1, 3]
        layer_strides = [1, 1]
        layer_paddings = [0, 1]

        return layer_kernel_sizes, layer_strides, layer_paddings


class _DenseBlock(nn.Sequential):
    def __init__(self, num_layers, num_input_features, bn_size, growth_rate, drop_rate):
        super(_DenseBlock, self).__init__()
        self.block_kernel_sizes = []
        self.block_strides = []
        self.block_paddings = []

        for i in range(num_layers):
            layer = _DenseLayer(num_input_features + i * growth_rate, growth_rate, bn_size, drop_rate)
            layer_kernel_sizes, layer_strides, layer_paddings = layer.layer_conv_info()
            self.block_kernel_sizes.extend(layer_kernel_sizes)
            self.block_strides.extend(layer_strides)
            self.block_paddings.extend(layer_paddings)
            self.add_module('denselayer%d' % (i + 1), layer)

        self.num_layers = _DenseLayer.num_layers * num_layers

    def block_conv_info(self):
        return self.block_kernel_sizes, self.block_strides, self.block_paddings


class _Transition(nn.Sequential):

    num_layers = 1

    def __init__(self, num_input_features, num_output_features):
        super(_Transition, self).__init__()
        self.add_module('norm', nn.BatchNorm2d(num_input_features))
        self.add_module('relu', nn.ReLU(inplace=True))
        self.add_module('conv', nn.Conv2d(num_input_features, num_output_features,
                                          kernel_size=1, stride=1, bias=False))
        self.add_module('pool', nn.AvgPool2d(kernel_size=2, stride=2)) # AvgPool2d has no padding

    def block_conv_info(self):
        return [1, 2], [1, 2], [0, 0]


class DenseNet_features(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
        num_init_features (int) - the number of filters to learn in the first convolution layer
        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
    """

    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):

        super(DenseNet_features, self).__init__()
        self.kernel_sizes = []
        self.strides = []
        self.paddings = []

        self.n_layers = 0

        # First convolution
        self.features = nn.Sequential(OrderedDict([
            ('conv0', nn.Conv2d(in_channels=3, out_channels=num_init_features, kernel_size=7, stride=2, padding=3, bias=False)),
            ('norm0', nn.BatchNorm2d(num_init_features)),
            ('relu0', nn.ReLU(inplace=True)),
            ('pool0', nn.MaxPool2d(kernel_size=3, stride=2, padding=1)),
        ]))

        self.kernel_sizes.extend([7, 3])
        self.strides.extend([2, 2])
        self.paddings.extend([3, 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, drop_rate=drop_rate)
            self.n_layers += block.num_layers

            block_kernel_sizes, block_strides, block_paddings = block.block_conv_info()
            self.kernel_sizes.extend(block_kernel_sizes)
            self.strides.extend(block_strides)
            self.paddings.extend(block_paddings)

            self.features.add_module('denseblock%d' % (i + 1), block)
            num_features = 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)

                self.n_layers += trans.num_layers

                block_kernel_sizes, block_strides, block_paddings = trans.block_conv_info()
                self.kernel_sizes.extend(block_kernel_sizes)
                self.strides.extend(block_strides)
                self.paddings.extend(block_paddings)

                self.features.add_module('transition%d' % (i + 1), trans)
                num_features = num_features // 2

        # Final batch norm
        self.features.add_module('norm5', nn.BatchNorm2d(num_features))
        self.features.add_module('final_relu', nn.ReLU(inplace=True))

        # Official init from torch repo.
        for m in self.modules():
            if isinstance(m, nn.Conv2d):
                nn.init.kaiming_normal_(m.weight)
            elif isinstance(m, nn.BatchNorm2d):
                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):
        return self.features(x)

    def conv_info(self):
        return self.kernel_sizes, self.strides, self.paddings

    def num_layers(self):
        return self.n_layers

    def __repr__(self):
        template = 'densenet{}_features'
        return template.format((self.num_layers() + 2))


def densenet121_features(pretrained=False, **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
    """
    model = DenseNet_features(num_init_features=64, growth_rate=32, block_config=(6, 12, 24, 16),
                     **kwargs)
    if pretrained:
        # '.'s are no longer allowed in module names, but pervious _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 = model_zoo.load_url(model_urls['densenet121'], model_dir=model_dir)
        for key in list(state_dict.keys()):
            '''
            example
            key 'features.denseblock4.denselayer24.norm.2.running_var'
            res.group(1) 'features.denseblock4.denselayer24.norm'
            res.group(2) '2.running_var'
            new_key 'features.denseblock4.denselayer24.norm2.running_var'
            '''
            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]

        del state_dict['classifier.weight']
        del state_dict['classifier.bias']
        model.load_state_dict(state_dict)
    return model


def densenet169_features(pretrained=False, **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
    """
    model = DenseNet_features(num_init_features=64, growth_rate=32, block_config=(6, 12, 32, 32),
                     **kwargs)
    if pretrained:
        # '.'s are no longer allowed in module names, but pervious _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 = model_zoo.load_url(model_urls['densenet169'], model_dir=model_dir)
        for key in list(state_dict.keys()):
            '''
            example
            key 'features.denseblock4.denselayer24.norm.2.running_var'
            res.group(1) 'features.denseblock4.denselayer24.norm'
            res.group(2) '2.running_var'
            new_key 'features.denseblock4.denselayer24.norm2.running_var'
            '''
            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]

        del state_dict['classifier.weight']
        del state_dict['classifier.bias']
        model.load_state_dict(state_dict)
    return model


def densenet201_features(pretrained=False, **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
    """
    model = DenseNet_features(num_init_features=64, growth_rate=32, block_config=(6, 12, 48, 32),
                     **kwargs)
    if pretrained:
        # '.'s are no longer allowed in module names, but pervious _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 = model_zoo.load_url(model_urls['densenet201'], model_dir=model_dir)
        for key in list(state_dict.keys()):
            '''
            example
            key 'features.denseblock4.denselayer24.norm.2.running_var'
            res.group(1) 'features.denseblock4.denselayer24.norm'
            res.group(2) '2.running_var'
            new_key 'features.denseblock4.denselayer24.norm2.running_var'
            '''
            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]

        del state_dict['classifier.weight']
        del state_dict['classifier.bias']
        model.load_state_dict(state_dict)

    return model


def densenet161_features(pretrained=False, **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
    """
    model = DenseNet_features(num_init_features=96, growth_rate=48, block_config=(6, 12, 36, 24),
                     **kwargs)
    if pretrained:
        # '.'s are no longer allowed in module names, but pervious _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 = model_zoo.load_url(model_urls['densenet161'], model_dir=model_dir)
        for key in list(state_dict.keys()):
            '''
            example
            key 'features.denseblock4.denselayer24.norm.2.running_var'
            res.group(1) 'features.denseblock4.denselayer24.norm'
            res.group(2) '2.running_var'
            new_key 'features.denseblock4.denselayer24.norm2.running_var'
            '''
            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]


        del state_dict['classifier.weight']
        del state_dict['classifier.bias']
        model.load_state_dict(state_dict)

    return model

if __name__ == '__main__':

    d161 = densenet161_features(True)
    print(d161)

    d201 = densenet201_features(True)
    print(d201)

    d169 = densenet169_features(True)
    print(d169)

    d121 = densenet121_features(True)
    print(d121)