摘要

这篇文章告诉大家如何在DBnet中加入新的主干网络。通过这篇文章你可以学到如何将现有的主干网络加入到DBNet中，提高DBNet的检测能力

主干网络

我加入的网络是ConvNext。代码详见：

# Copyright (c) Meta Platforms, Inc. and affiliates.

# All rights reserved.

# This source code is licensed under the license found in the
# LICENSE file in the root directory of this source tree.


import torch
import torch.nn as nn
import torch.nn.functional as F
from timm.models.layers import trunc_normal_, DropPath
from timm.models.registry import register_model

class Block(nn.Module):
    r""" ConvNeXt Block. There are two equivalent implementations:
    (1) DwConv -> LayerNorm (channels_first) -> 1x1 Conv -> GELU -> 1x1 Conv; all in (N, C, H, W)
    (2) DwConv -> Permute to (N, H, W, C); LayerNorm (channels_last) -> Linear -> GELU -> Linear; Permute back
    We use (2) as we find it slightly faster in PyTorch
    
    Args:
        dim (int): Number of input channels.
        drop_path (float): Stochastic depth rate. Default: 0.0
        layer_scale_init_value (float): Init value for Layer Scale. Default: 1e-6.
    """
    def __init__(self, dim, drop_path=0., layer_scale_init_value=1e-6):
        super().__init__()
        self.dwconv = nn.Conv2d(dim, dim, kernel_size=7, padding=3, groups=dim) # depthwise conv
        self.norm = LayerNorm(dim, eps=1e-6)
        self.pwconv1 = nn.Linear(dim, 4 * dim) # pointwise/1x1 convs, implemented with linear layers
        self.act = nn.GELU()
        self.pwconv2 = nn.Linear(4 * dim, dim)
        self.gamma = nn.Parameter(layer_scale_init_value * torch.ones((dim)), 
                                    requires_grad=True) if layer_scale_init_value > 0 else None
        self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()

    def forward(self, x):
        input = x
        x = self.dwconv(x)
        x = x.permute(0, 2, 3, 1) # (N, C, H, W) -> (N, H, W, C)
        x = self.norm(x)
        x = self.pwconv1(x)
        x = self.act(x)
        x = self.pwconv2(x)
        if self.gamma is not None:
            x = self.gamma * x
        x = x.permute(0, 3, 1, 2) # (N, H, W, C) -> (N, C, H, W)

        x = input + self.drop_path(x)
        return x

class ConvNeXt(nn.Module):
    r""" ConvNeXt
        A PyTorch impl of : `A ConvNet for the 2020s`  -
          https://arxiv.org/pdf/2201.03545.pdf

    Args:
        in_chans (int): Number of input image channels. Default: 3
        num_classes (int): Number of classes for classification head. Default: 1000
        depths (tuple(int)): Number of blocks at each stage. Default: [3, 3, 9, 3]
        dims (int): Feature dimension at each stage. Default: [96, 192, 384, 768]
        drop_path_rate (float): Stochastic depth rate. Default: 0.
        layer_scale_init_value (float): Init value for Layer Scale. Default: 1e-6.
        head_init_scale (float): Init scaling value for classifier weights and biases. Default: 1.
    """
    def __init__(self, in_chans=3, num_classes=1000, 
                 depths=[3, 3, 9, 3], dims=[96, 192, 384, 768], drop_path_rate=0., 
                 layer_scale_init_value=1e-6, head_init_scale=1.,
                 ):
        super().__init__()

        self.downsample_layers = nn.ModuleList() # stem and 3 intermediate downsampling conv layers
        stem = nn.Sequential(
            nn.Conv2d(in_chans, dims[0], kernel_size=4, stride=4),
            LayerNorm(dims[0], eps=1e-6, data_format="channels_first")
        )
        self.downsample_layers.append(stem)
        for i in range(3):
            downsample_layer = nn.Sequential(
                    LayerNorm(dims[i], eps=1e-6, data_format="channels_first"),
                    nn.Conv2d(dims[i], dims[i+1], kernel_size=2, stride=2),
            )
            self.downsample_layers.append(downsample_layer)

        self.stages = nn.ModuleList() # 4 feature resolution stages, each consisting of multiple residual blocks
        dp_rates=[x.item() for x in torch.linspace(0, drop_path_rate, sum(depths))] 
        cur = 0
        for i in range(4):
            stage = nn.Sequential(
                *[Block(dim=dims[i], drop_path=dp_rates[cur + j], 
                layer_scale_init_value=layer_scale_init_value) for j in range(depths[i])]
            )
            self.stages.append(stage)
            cur += depths[i]

        self.norm = nn.LayerNorm(dims[-1], eps=1e-6) # final norm layer
        self.head = nn.Linear(dims[-1], num_classes)

        self.apply(self._init_weights)
        self.head.weight.data.mul_(head_init_scale)
        self.head.bias.data.mul_(head_init_scale)

    def _init_weights(self, m):
        if isinstance(m, (nn.Conv2d, nn.Linear)):
            trunc_normal_(m.weight, std=.02)
            nn.init.constant_(m.bias, 0)

    def forward_features(self, x):
        for i in range(4):
            x = self.downsample_layers[i](x)
            x = self.stages[i](x)
        return self.norm(x.mean([-2, -1])) # global average pooling, (N, C, H, W) -> (N, C)

    def forward(self, x):
        x = self.forward_features(x)
        x = self.head(x)
        return x

class LayerNorm(nn.Module):
    r""" LayerNorm that supports two data formats: channels_last (default) or channels_first. 
    The ordering of the dimensions in the inputs. channels_last corresponds to inputs with 
    shape (batch_size, height, width, channels) while channels_first corresponds to inputs 
    with shape (batch_size, channels, height, width).
    """
    def __init__(self, normalized_shape, eps=1e-6, data_format="channels_last"):
        super().__init__()
        self.weight = nn.Parameter(torch.ones(normalized_shape))
        self.bias = nn.Parameter(torch.zeros(normalized_shape))
        self.eps = eps
        self.data_format = data_format
        if self.data_format not in ["channels_last", "channels_first"]:
            raise NotImplementedError 
        self.normalized_shape = (normalized_shape, )
    
    def forward(self, x):
        if self.data_format == "channels_last":
            return F.layer_norm(x, self.normalized_shape, self.weight, self.bias, self.eps)
        elif self.data_format == "channels_first":
            u = x.mean(1, keepdim=True)
            s = (x - u).pow(2).mean(1, keepdim=True)
            x = (x - u) / torch.sqrt(s + self.eps)
            x = self.weight[:, None, None] * x + self.bias[:, None, None]
            return x


model_urls = {
    "convnext_tiny_1k": "https://dl.fbaipublicfiles.com/convnext/convnext_tiny_1k_224_ema.pth",
    "convnext_small_1k": "https://dl.fbaipublicfiles.com/convnext/convnext_small_1k_224_ema.pth",
    "convnext_base_1k": "https://dl.fbaipublicfiles.com/convnext/convnext_base_1k_224_ema.pth",
    "convnext_large_1k": "https://dl.fbaipublicfiles.com/convnext/convnext_large_1k_224_ema.pth",
    "convnext_base_22k": "https://dl.fbaipublicfiles.com/convnext/convnext_base_22k_224.pth",
    "convnext_large_22k": "https://dl.fbaipublicfiles.com/convnext/convnext_large_22k_224.pth",
    "convnext_xlarge_22k": "https://dl.fbaipublicfiles.com/convnext/convnext_xlarge_22k_224.pth",
}

@register_model
def convnext_tiny(pretrained=False, **kwargs):
    model = ConvNeXt(depths=[3, 3, 9, 3], dims=[96, 192, 384, 768], **kwargs)
    if pretrained:
        url = model_urls['convnext_tiny_1k']
        checkpoint = torch.hub.load_state_dict_from_url(url=url, map_location="cpu", check_hash=True)
        model.load_state_dict(checkpoint["model"])
    return model

@register_model
def convnext_small(pretrained=False, **kwargs):
    model = ConvNeXt(depths=[3, 3, 27, 3], dims=[96, 192, 384, 768], **kwargs)
    if pretrained:
        url = model_urls['convnext_small_1k']
        checkpoint = torch.hub.load_state_dict_from_url(url=url, map_location="cpu", check_hash=True)
        model.load_state_dict(checkpoint["model"])
    return model

@register_model
def convnext_base(pretrained=False, in_22k=False, **kwargs):
    model = ConvNeXt(depths=[3, 3, 27, 3], dims=[128, 256, 512, 1024], **kwargs)
    if pretrained:
        url = model_urls['convnext_base_22k'] if in_22k else model_urls['convnext_base_1k']
        checkpoint = torch.hub.load_state_dict_from_url(url=url, map_location="cpu", check_hash=True)
        model.load_state_dict(checkpoint["model"])
    return model

@register_model
def convnext_large(pretrained=False, in_22k=False, **kwargs):
    model = ConvNeXt(depths=[3, 3, 27, 3], dims=[192, 384, 768, 1536], **kwargs)
    if pretrained:
        url = model_urls['convnext_large_22k'] if in_22k else model_urls['convnext_large_1k']
        checkpoint = torch.hub.load_state_dict_from_url(url=url, map_location="cpu", check_hash=True)
        model.load_state_dict(checkpoint["model"])
    return model

@register_model
def convnext_xlarge(pretrained=False, in_22k=False, **kwargs):
    model = ConvNeXt(depths=[3, 3, 27, 3], dims=[256, 512, 1024, 2048], **kwargs)
    if pretrained:
        url = model_urls['convnext_xlarge_22k'] if in_22k else model_urls['convnext_xlarge_1k']
        checkpoint = torch.hub.load_state_dict_from_url(url=url, map_location="cpu", check_hash=True)
        model.load_state_dict(checkpoint["model"])
    return model

操作过程

修改主干网络

我们把上面的代码改造成DBNet的主干网络。
DBNet的代码：https://github.com/WenmuZhou/DBNet.pytorch
将其下载下来。
然后在./models/backbone下面新建convnext.py脚本。
将上面的代码，插入进来。
然后修改ConvNeXt的 forward()函数，修改为：

 def forward(self, x):
        x= self.downsample_layers[0](x)
        x2 = self.stages[0](x)

        x = self.downsample_layers[1](x2)
        x3 = self.stages[1](x)

        x = self.downsample_layers[2](x3)
        x4 = self.stages[2](x)

        x = self.downsample_layers[3](x4)
        x5 = self.stages[3](x)

        return x2,x3,x4,x5

和网络图对应起来。

在 def __init__函数中 增加out_channels属性

 self.out_channels=dims

将in_chans改为in_channels，这样就能和配置文件的字段对上了。

删除@register_model

接下来修改创建模型的部分，用convnext_tiny举例

def convnext_tiny(pretrained=False, **kwargs):
    model = ConvNeXt(depths=[2, 2, 3, 3], dims=[96, 192, 384, 768], **kwargs)
    if pretrained:
        url = model_urls['convnext_tiny_1k']
        checkpoint = torch.hub.load_state_dict_from_url(url=url, map_location="cpu", check_hash=True)
        model.load_state_dict(checkpoint["model"],strict=False)
    return model

原始的convnext_tiny很大，我在这里把depths做了修改，
将depths=[3, 3, 9, 3]改为depths=[2, 2, 3, 3]，设置strict为False，防止预训练权重对不上报错。
在代码的开始部分增加：

__all__ = ['convnext_tiny', 'convnext_small', 'convnext_base']

这样可以访问的方法就这三个了。

修改完成后测一下：

if __name__ == '__main__':
    import torch

    x = torch.zeros(2, 3, 640, 640)
    net = convnext_tiny(pretrained=False)
    y = net(x)
    for u in y:
        print(u.shape)

    print(net.out_channels)

完整代码：

# Copyright (c) Meta Platforms, Inc. and affiliates.

# All rights reserved.

# This source code is licensed under the license found in the
# LICENSE file in the root directory of this source tree.


import torch
import torch.nn as nn
import torch.nn.functional as F
from timm.models.layers import trunc_normal_, DropPath

__all__ = ['convnext_tiny', 'convnext_small', 'convnext_base']


class Block(nn.Module):
    r""" ConvNeXt Block. There are two equivalent implementations:
    (1) DwConv -> LayerNorm (channels_first) -> 1x1 Conv -> GELU -> 1x1 Conv; all in (N, C, H, W)
    (2) DwConv -> Permute to (N, H, W, C); LayerNorm (channels_last) -> Linear -> GELU -> Linear; Permute back
    We use (2) as we find it slightly faster in PyTorch

    Args:
        dim (int): Number of input channels.
        drop_path (float): Stochastic depth rate. Default: 0.0
        layer_scale_init_value (float): Init value for Layer Scale. Default: 1e-6.
    """

    def __init__(self, dim, drop_path=0., layer_scale_init_value=1e-6):
        super().__init__()
        self.dwconv = nn.Conv2d(dim, dim, kernel_size=7, padding=3, groups=dim)  # depthwise conv
        self.norm = LayerNorm(dim, eps=1e-6)
        self.pwconv1 = nn.Linear(dim, 4 * dim)  # pointwise/1x1 convs, implemented with linear layers
        self.act = nn.GELU()
        self.pwconv2 = nn.Linear(4 * dim, dim)
        self.gamma = nn.Parameter(layer_scale_init_value * torch.ones((dim)),
                                  requires_grad=True) if layer_scale_init_value > 0 else None
        self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()

    def forward(self, x):
        input = x
        x = self.dwconv(x)
        x = x.permute(0, 2, 3, 1)  # (N, C, H, W) -> (N, H, W, C)
        x = self.norm(x)
        x = self.pwconv1(x)
        x = self.act(x)
        x = self.pwconv2(x)
        if self.gamma is not None:
            x = self.gamma * x
        x = x.permute(0, 3, 1, 2)  # (N, H, W, C) -> (N, C, H, W)

        x = input + self.drop_path(x)
        return x


class ConvNeXt(nn.Module):
    r""" ConvNeXt
        A PyTorch impl of : `A ConvNet for the 2020s`  -
          https://arxiv.org/pdf/2201.03545.pdf

    Args:
        in_chans (int): Number of input image channels. Default: 3
        num_classes (int): Number of classes for classification head. Default: 1000
        depths (tuple(int)): Number of blocks at each stage. Default: [3, 3, 9, 3]
        dims (int): Feature dimension at each stage. Default: [96, 192, 384, 768]
        drop_path_rate (float): Stochastic depth rate. Default: 0.
        layer_scale_init_value (float): Init value for Layer Scale. Default: 1e-6.
        head_init_scale (float): Init scaling value for classifier weights and biases. Default: 1.
    """

    def __init__(self, in_channels=3, num_classes=1000,
                 depths=[3, 3, 3, 3], dims=[96, 192, 384, 768], drop_path_rate=0.,
                 layer_scale_init_value=1e-6, head_init_scale=1.,
                 ):
        super().__init__()

        self.downsample_layers = nn.ModuleList()  # stem and 3 intermediate downsampling conv layers
        stem = nn.Sequential(
            nn.Conv2d(in_channels, dims[0], kernel_size=4, stride=4),
            LayerNorm(dims[0], eps=1e-6, data_format="channels_first")
        )
        self.downsample_layers.append(stem)
        for i in range(3):
            downsample_layer = nn.Sequential(
                LayerNorm(dims[i], eps=1e-6, data_format="channels_first"),
                nn.Conv2d(dims[i], dims[i + 1], kernel_size=2, stride=2),
            )
            self.downsample_layers.append(downsample_layer)

        self.stages = nn.ModuleList()  # 4 feature resolution stages, each consisting of multiple residual blocks
        dp_rates = [x.item() for x in torch.linspace(0, drop_path_rate, sum(depths))]
        cur = 0
        for i in range(4):
            stage = nn.Sequential(
                *[Block(dim=dims[i], drop_path=dp_rates[cur + j],
                        layer_scale_init_value=layer_scale_init_value) for j in range(depths[i])]
            )
            self.stages.append(stage)
            cur += depths[i]

        self.norm = nn.LayerNorm(dims[-1], eps=1e-6)  # final norm layer
        self.head = nn.Linear(dims[-1], num_classes)

        self.apply(self._init_weights)
        self.head.weight.data.mul_(head_init_scale)
        self.head.bias.data.mul_(head_init_scale)
        self.out_channels=[96,192,384,768]

    def _init_weights(self, m):
        if isinstance(m, (nn.Conv2d, nn.Linear)):
            trunc_normal_(m.weight, std=.02)
            nn.init.constant_(m.bias, 0)

    # def forward_features(self, x):
    #     for i in range(4):
    #         x = self.downsample_layers[i](x)
    #         x = self.stages[i](x)
    #     return self.norm(x.mean([-2, -1]))  # global average pooling, (N, C, H, W) -> (N, C)

    def forward(self, x):
        x= self.downsample_layers[0](x)
        x2 = self.stages[0](x)

        x = self.downsample_layers[1](x2)
        x3 = self.stages[1](x)

        x = self.downsample_layers[2](x3)
        x4 = self.stages[2](x)

        x = self.downsample_layers[3](x4)
        x5 = self.stages[3](x)

        return x2,x3,x4,x5


class LayerNorm(nn.Module):
    r""" LayerNorm that supports two data formats: channels_last (default) or channels_first.
    The ordering of the dimensions in the inputs. channels_last corresponds to inputs with
    shape (batch_size, height, width, channels) while channels_first corresponds to inputs
    with shape (batch_size, channels, height, width).
    """

    def __init__(self, normalized_shape, eps=1e-6, data_format="channels_last"):
        super().__init__()
        self.weight = nn.Parameter(torch.ones(normalized_shape))
        self.bias = nn.Parameter(torch.zeros(normalized_shape))
        self.eps = eps
        self.data_format = data_format
        if self.data_format not in ["channels_last", "channels_first"]:
            raise NotImplementedError
        self.normalized_shape = (normalized_shape,)

    def forward(self, x):
        if self.data_format == "channels_last":
            return F.layer_norm(x, self.normalized_shape, self.weight, self.bias, self.eps)
        elif self.data_format == "channels_first":
            u = x.mean(1, keepdim=True)
            s = (x - u).pow(2).mean(1, keepdim=True)
            x = (x - u) / torch.sqrt(s + self.eps)
            x = self.weight[:, None, None] * x + self.bias[:, None, None]
            return x


model_urls = {
    "convnext_tiny_1k": "https://dl.fbaipublicfiles.com/convnext/convnext_tiny_1k_224_ema.pth",
    "convnext_small_1k": "https://dl.fbaipublicfiles.com/convnext/convnext_small_1k_224_ema.pth",
    "convnext_base_1k": "https://dl.fbaipublicfiles.com/convnext/convnext_base_1k_224_ema.pth",
    "convnext_large_1k": "https://dl.fbaipublicfiles.com/convnext/convnext_large_1k_224_ema.pth",
    "convnext_base_22k": "https://dl.fbaipublicfiles.com/convnext/convnext_base_22k_224.pth",
    "convnext_large_22k": "https://dl.fbaipublicfiles.com/convnext/convnext_large_22k_224.pth",
    "convnext_xlarge_22k": "https://dl.fbaipublicfiles.com/convnext/convnext_xlarge_22k_224.pth",
}


def convnext_tiny(pretrained=False, **kwargs):
    model = ConvNeXt(depths=[2, 2, 3, 3], dims=[96, 192, 384, 768], **kwargs)
    if pretrained:
        url = model_urls['convnext_tiny_1k']
        checkpoint = torch.hub.load_state_dict_from_url(url=url, map_location="cpu", check_hash=True)
        model.load_state_dict(checkpoint["model"],strict=False)
    return model


def convnext_small(pretrained=False, **kwargs):
    model = ConvNeXt(depths=[3, 3, 27, 3], dims=[96, 192, 384, 768], **kwargs)
    if pretrained:
        url = model_urls['convnext_small_1k']
        checkpoint = torch.hub.load_state_dict_from_url(url=url, map_location="cpu", check_hash=True)
        model.load_state_dict(checkpoint["model"])
    return model


def convnext_base(pretrained=False, in_22k=False, **kwargs):
    model = ConvNeXt(depths=[3, 3, 27, 3], dims=[128, 256, 512, 1024], **kwargs)
    if pretrained:
        url = model_urls['convnext_base_22k'] if in_22k else model_urls['convnext_base_1k']
        checkpoint = torch.hub.load_state_dict_from_url(url=url, map_location="cpu", check_hash=True)
        model.load_state_dict(checkpoint["model"])
    return model


def convnext_large(pretrained=False, in_22k=False, **kwargs):
    model = ConvNeXt(depths=[3, 3, 27, 3], dims=[192, 384, 768, 1536], **kwargs)
    if pretrained:
        url = model_urls['convnext_large_22k'] if in_22k else model_urls['convnext_large_1k']
        checkpoint = torch.hub.load_state_dict_from_url(url=url, map_location="cpu", check_hash=True)
        model.load_state_dict(checkpoint["model"])
    return model


def convnext_xlarge(pretrained=False, in_22k=False, **kwargs):
    model = ConvNeXt(depths=[3, 3, 27, 3], dims=[256, 512, 1024, 2048], **kwargs)
    if pretrained:
        url = model_urls['convnext_xlarge_22k'] if in_22k else model_urls['convnext_xlarge_1k']
        checkpoint = torch.hub.load_state_dict_from_url(url=url, map_location="cpu", check_hash=True)
        model.load_state_dict(checkpoint["model"])
    return model


if __name__ == '__main__':
    import torch

    x = torch.zeros(2, 3, 640, 640)
    net = convnext_tiny(pretrained=False)
    y = net(x)
    for u in y:
        print(u.shape)

    print(net.out_channels)

将主干网络加到配置列表

修改完成后，在__init__.py导入convnext，在support_backbone列表中增加convnext_tiny字段，如下图：

接下来在./config文件夹中增加convnext的配置文件，命名为：icdar2015_convnext_FPN_DBhead_polyLR.yaml
从其他的配置文件中复制一份参数，
然后将backbone的type修改为 convnext_tiny

完整代码：

name: DBNet
base: ['config/icdar2015.yaml']
arch:
  type: Model
  backbone:
    type: convnext_tiny
    pretrained: true
  neck:
    type: FPN
    inner_channels: 384
  head:
    type: DBHead
    out_channels: 2
    k: 50
post_processing:
  type: SegDetectorRepresenter
  args:
    thresh: 0.3
    box_thresh: 0.7
    max_candidates: 1000
    unclip_ratio: 1.5 # from paper
metric:
  type: QuadMetric
  args:
    is_output_polygon: false
loss:
  type: DBLoss
  alpha: 1
  beta: 10
  ohem_ratio: 3
optimizer:
  type: Adam
  args:
    lr: 0.001
    weight_decay: 0
    amsgrad: true
lr_scheduler:
  type: WarmupPolyLR
  args:
    warmup_epoch: 3
trainer:
  seed: 2
  epochs: 1200
  log_iter: 10
  show_images_iter: 50
  resume_checkpoint: ''
  finetune_checkpoint: ''
  output_dir: output
  tensorboard: true
dataset:
  train:
    dataset:
      args:
        data_path:
          - ./datasets/train.txt
        img_mode: RGB
    loader:
      batch_size: 2
      shuffle: true
      pin_memory: true
      num_workers: 1
      collate_fn: ''
  validate:
    dataset:
      args:
        data_path:
          - ./datasets/test.txt
        pre_processes:
          - type: ResizeShortSize
            args:
              short_size: 736
              resize_text_polys: false
        img_mode: RGB
    loader:
      batch_size: 1
      shuffle: true
      pin_memory: false
      num_workers: 1
      collate_fn: ICDARCollectFN

然后就可以运行了。