jittor.models.resnet 源代码

# ***************************************************************
# Copyright (c) 2023 Jittor. All Rights Reserved. 
# Maintainers: 
#     Guowei Yang <471184555@qq.com>
#     Wenyang Zhou <576825820@qq.com>
#     Dun Liang <randonlang@gmail.com>. 
# 
# This file is subject to the terms and conditions defined in
# file 'LICENSE.txt', which is part of this source code package.
# ***************************************************************
# This model is generated by pytorch converter.
import jittor as jt
from jittor import nn

__all__ = ['ResNet', 'Resnet18', 'Resnet34', 'Resnet26', 'Resnet38', 'Resnet50', 'Resnet101', 'Resnet152', 'Resnext50_32x4d', 'Resnext101_32x8d', 'Wide_resnet50_2', 'Wide_resnet101_2',
    'resnet18', 'resnet34', 'resnet26', 'resnet38', 'resnet50', 'resnet101', 'resnet152', 'resnext50_32x4d', 'resnext101_32x8d', 'wide_resnet50_2', 'wide_resnet101_2']

def conv3x3(in_planes, out_planes, stride=1, groups=1, dilation=1):
    conv=nn.Conv(in_planes, out_planes, kernel_size=3, stride=stride, padding=dilation, groups=groups, bias=False, dilation=dilation)
    jt.init.relu_invariant_gauss_(conv.weight, mode="fan_out")
    return conv

def conv1x1(in_planes, out_planes, stride=1):
    conv=nn.Conv(in_planes, out_planes, kernel_size=1, stride=stride, bias=False)
    jt.init.relu_invariant_gauss_(conv.weight, mode="fan_out")
    return conv

class BasicBlock(nn.Module):
    expansion = 1

    def __init__(self, inplanes, planes, stride=1, downsample=None, groups=1, base_width=64, dilation=1, norm_layer=None):
        super(BasicBlock, self).__init__()
        if (norm_layer is None):
            norm_layer = nn.BatchNorm
        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')
        self.conv1 = conv3x3(inplanes, planes, stride)
        self.bn1 = norm_layer(planes)
        self.relu = nn.Relu()
        self.conv2 = conv3x3(planes, planes)
        self.bn2 = norm_layer(planes)
        self.downsample = downsample
        self.stride = stride

    def execute(self, x):
        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):
    expansion = 4

    def __init__(self, inplanes, planes, stride=1, downsample=None, groups=1, base_width=64, dilation=1, norm_layer=None):
        super(Bottleneck, self).__init__()
        if (norm_layer is None):
            norm_layer = nn.BatchNorm
        width = (int((planes * (base_width / 64.0))) * groups)
        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()
        self.downsample = downsample
        self.stride = stride

    def execute(self, x):
        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

[文档] class ResNet(nn.Module): ''' ResNet源自论文 `Deep Residual Learning for Image Recognition <https://arxiv.org/abs/1512.03385>`__, 其通过添加shortcut来改善梯度回传, 允许更深层次的网络有效训练。 参数: - block (nn.Module): 残差块的类, 用于建立 ResNet 的不同层。 `block` 参数通常是一个实现了特定类型残差块结构的类, 比如 'BasicBlock' 或 'Bottleneck'。 - layers (list[int]): 每个 ResNet 层中残差块的数量, 是一个由4个元素组成的列表 - num_classes (int, optional): 分类任务中的类别数。默认值: 1000 - zero_init_residual (bool, optional): 是否使用零初始化残差块。默认值: False - groups (int, optional): 分组卷积的组数量。1意味着不使用分组卷积。默认值: 1 - width_per_group (int, optional): 每个分组的宽度。默认值: 64 - replace_stride_with_dilation (list[bool], optional): 是否用空洞卷积替换某些层的步长。若不为None则应为包含三个布尔值的列表。默认值: None - norm_layer (nn.Module, optional): 使用的标准化层, 默认使用 'nn.BatchNorm'。默认值: None 属性: - conv1 (nn.Conv): ResNet 的第一个卷积层。 - bn1 (nn.BatchNorm): ResNet 的第一个标准化层。 - relu (nn.Relu): Relu激活函数。 - maxpool (nn.Pool): 最大池化层。 - layer1 (nn.Sequential): ResNet 的第一层。 - layer2 (nn.Sequential): ResNet 的第二层。 - layer3 (nn.Sequential): ResNet 的第三层。 - layer4 (nn.Sequential): ResNet 的第四层。 - avgpool (nn.AdaptiveAvgPool2d): 自适应平均池化层。 - fc (nn.Linear): 全连接层。 代码示例: >>> import jittor as jt >>> from jittor.models.resnet import Bottleneck, ResNet >>> resnet = ResNet(Bottleneck, [3,4,6,3]) >>> print(resnet) ResNet( conv1: Conv(3, 64, (7, 7), (2, 2), (3, 3), (1, 1), 1, None, None, Kw=None, fan=None, i=None, bound=None) ...) ''' def __init__(self, block, layers, num_classes=1000, zero_init_residual=False, groups=1, width_per_group=64, replace_stride_with_dilation=None, norm_layer=None): super(ResNet, self).__init__() if (norm_layer is None): norm_layer = nn.BatchNorm self._norm_layer = norm_layer self.inplanes = 64 self.dilation = 1 if (replace_stride_with_dilation is None): replace_stride_with_dilation = [False, False, False] if (len(replace_stride_with_dilation) != 3): raise ValueError('replace_stride_with_dilation should be None or a 3-element tuple, got {}'.format(replace_stride_with_dilation)) self.groups = groups self.base_width = width_per_group self.conv1 = nn.Conv(3, self.inplanes, kernel_size=7, stride=2, padding=3, bias=False) jt.init.relu_invariant_gauss_(self.conv1.weight, mode="fan_out") self.bn1 = norm_layer(self.inplanes) self.relu = nn.Relu() self.maxpool = nn.Pool(kernel_size=3, stride=2, padding=1, op='maximum') 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) def _make_layer(self, block, planes, blocks, stride=1, dilate=False): 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): 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).float_auto() x = jt.reshape(x, (x.shape[0], -1)) x = self.fc(x) return x def execute(self, x): return self._forward_impl(x)
def _resnet(block, layers, **kwargs): model = ResNet(block, layers, **kwargs) return model
[文档] def Resnet18(pretrained=False, **kwargs): ''' 构建Resnet18模型 ResNet18模型源自论文 `Deep Residual Learning for Image Recognition <https://arxiv.org/abs/1512.03385>`__, 其通过添加shortcut来改善梯度回传, 允许更深层次的网络有效训练。 参数: - `pretrained` (bool): 表示是否加载预训练的ResNet18模型。默认为 `False`。如果设为 `True`, 函数会自动下载并加载预训练的ResNet18模型。 - `**kwargs`: 可变参数, 允许用户传递额外的、自定义的参数给 `_resnet` 函数。 返回值: - 返回一个ResNet18模型实例。如果 `pretrained=True`, 则返回的模型将加载预训练权重;否则, 返回一个未经训练的ResNet18模型。 代码示例: >>> import jittor as jt >>> from jittor.models.resnet import * >>> net = Resnet18(pretrained=True) >>> x = jt.rand(1, 3, 224, 224) >>> y = net(x) >>> y.shape [1, 1000] ''' model = _resnet(BasicBlock, [2, 2, 2, 2], **kwargs) if pretrained: model.load("jittorhub://resnet18.pkl") return model
resnet18 = Resnet18
[文档] def Resnet34(pretrained=False, **kwargs): ''' 构建Resnet34模型 ResNet34模型源自论文 `Deep Residual Learning for Image Recognition <https://arxiv.org/abs/1512.03385>`__, 其通过添加shortcut来改善梯度回传, 允许更深层次的网络有效训练。 参数: - `pretrained` (bool): 表示是否加载预训练的ResNet34模型。默认为 `False`。如果设为 `True`, 函数会自动下载并加载预训练的ResNet34模型。 - `**kwargs`: 可变参数, 允许用户传递额外的、自定义的参数给 `_resnet` 函数。 返回值: - 返回一个ResNet34模型实例。如果 `pretrained=True`, 则返回的模型将加载预训练权重;否则, 返回一个未经训练的ResNet34模型。 代码示例: >>> import jittor as jt >>> from jittor.models.resnet import * >>> net = Resnet34(pretrained=True) >>> x = jt.rand(1, 3, 224, 224) >>> y = net(x) >>> y.shape [1, 1000] ''' model = _resnet(BasicBlock, [3, 4, 6, 3], **kwargs) if pretrained: model.load("jittorhub://resnet34.pkl") return model
resnet34 = Resnet34
[文档] def Resnet50(pretrained=False, **kwargs): ''' 构建Resnet50模型 ResNet50模型源自论文 `Deep Residual Learning for Image Recognition <https://arxiv.org/abs/1512.03385>`__, 其通过添加shortcut来改善梯度回传, 允许更深层次的网络有效训练。 参数: - `pretrained` (bool): 表示是否加载预训练的ResNet50模型。默认为 `False`。如果设为 `True`, 函数会自动下载并加载预训练的ResNet50模型。 - `**kwargs`: 可变参数, 允许用户传递额外的、自定义的参数给 `_resnet` 函数。 返回值: - 返回一个ResNet50模型实例。如果 `pretrained=True`, 则返回的模型将加载预训练权重;否则, 返回一个未经训练的ResNet50模型。 代码示例: >>> import jittor as jt >>> from jittor.models.resnet import * >>> net = Resnet50(pretrained=True) >>> x = jt.rand(1, 3, 224, 224) >>> y = net(x) >>> y.shape [1, 1000] ''' model = _resnet(Bottleneck, [3, 4, 6, 3], **kwargs) if pretrained: model.load("jittorhub://resnet50.pkl") return model
resnet50 = Resnet50
[文档] def Resnet38(pretrained=False, **kwargs): ''' 构建ResNet38模型 ResNet38模型源自论文 `Deep Residual Learning for Image Recognition <https://arxiv.org/abs/1512.03385>`__, 其通过添加shortcut来改善梯度回传, 允许更深层次的网络有效训练。 参数: - `pretrained` (bool): 表示是否加载预训练的ResNet38模型。默认为 `False`。如果设为 `True`, 函数会自动下载并加载预训练的ResNet38模型。 - `**kwargs`: 可变参数, 允许用户传递额外的、自定义的参数给 `_resnet` 函数。 返回值: - 返回一个ResNet38模型实例。如果 `pretrained=True`, 则返回的模型将加载预训练权重;否则, 返回一个未经训练的ResNet38模型。 代码示例: >>> import jittor as jt >>> from jittor.models.resnet import * >>> net = ResNet38(pretrained=True) >>> x = jt.rand(1, 3, 224, 224) >>> y = net(x) >>> y.shape [1, 1000] ''' model = _resnet(Bottleneck, [2, 3, 5, 2], **kwargs) if pretrained: model.load("jittorhub://resnet38.pkl") return model
resnet38 = Resnet38
[文档] def Resnet26(pretrained=False, **kwargs): ''' 构建ResNet26模型 ResNet26模型源自论文 `Deep Residual Learning for Image Recognition <https://arxiv.org/abs/1512.03385>`__, 其通过添加shortcut来改善梯度回传, 允许更深层次的网络有效训练。 参数: - `pretrained` (bool): 表示是否加载预训练的ResNet26模型。默认为 `False`。如果设为 `True`, 函数会自动下载并加载预训练的ResNet26模型。 - `**kwargs`: 可变参数, 允许用户传递额外的、自定义的参数给 `_resnet` 函数。 返回值: - 返回一个ResNet26模型实例。如果 `pretrained=True`, 则返回的模型将加载预训练权重;否则, 返回一个未经训练的ResNet26模型。 代码示例: >>> import jittor as jt >>> from jittor.models.resnet import * >>> net = ResNet26(pretrained=True) >>> x = jt.rand(1, 3, 224, 224) >>> y = net(x) >>> y.shape [1, 1000] ''' model = _resnet(Bottleneck, [1, 2, 4, 1], **kwargs) if pretrained: model.load("jittorhub://resnet26.pkl") return model
resnet26 = Resnet26
[文档] def Resnet101(pretrained=False, **kwargs): ''' 构建ResNet101模型 ResNet101模型源自论文 `Deep Residual Learning for Image Recognition <https://arxiv.org/abs/1512.03385>`__, 其通过添加shortcut来改善梯度回传, 允许更深层次的网络有效训练。 参数: - `pretrained` (bool): 表示是否加载预训练的ResNet101模型。默认为 `False`。如果设为 `True`, 函数会自动下载并加载预训练的ResNet101模型。 - `**kwargs`: 可变参数, 允许用户传递额外的、自定义的参数给 `_resnet` 函数。 返回值: - 返回一个ResNet101模型实例。如果 `pretrained=True`, 则返回的模型将加载预训练权重;否则, 返回一个未经训练的ResNet101模型。 代码示例: >>> import jittor as jt >>> from jittor.models.resnet import * >>> net = ResNet101(pretrained=True) >>> x = jt.rand(1, 3, 224, 224) >>> y = net(x) >>> y.shape [1, 1000] ''' model = _resnet(Bottleneck, [3, 4, 23, 3], **kwargs) if pretrained: model.load("jittorhub://resnet101.pkl") return model
resnet101 = Resnet101
[文档] def Resnet152(pretrained=False, **kwargs): ''' 构建ResNet152模型 ResNet152模型源自论文 `Deep Residual Learning for Image Recognition <https://arxiv.org/abs/1512.03385>`__, 其通过添加shortcut来改善梯度回传, 允许更深层次的网络有效训练。 参数: - `pretrained` (bool): 表示是否加载预训练的ResNet152模型。默认为 `False`。如果设为 `True`, 函数会自动下载并加载预训练的ResNet152模型。 - `**kwargs`: 可变参数, 允许用户传递额外的、自定义的参数给 `_resnet` 函数。 返回值: - 返回一个ResNet152模型实例。如果 `pretrained=True`, 则返回的模型将加载预训练权重;否则, 返回一个未经训练的ResNet152模型。 代码示例: >>> import jittor as jt >>> from jittor.models.resnet import * >>> net = ResNet152(pretrained=True) >>> x = jt.rand(1, 3, 224, 224) >>> y = net(x) >>> y.shape [1, 1000] ''' model = _resnet(Bottleneck, [3, 8, 36, 3], **kwargs) if pretrained: model.load("jittorhub://resnet152.pkl") return model
resnet152 = Resnet152
[文档] def Resnext50_32x4d(pretrained=False, **kwargs): ''' 构建一个Resnext50_32x4d模型 Resnext源自论文 `Aggregated Residual Transformations for Deep Neural Networks <https://arxiv.org/abs/1611.05431>`__。 参数: - `pretrained` (bool, optional): 表示是否预加载预训练模型。默认为 `False`。 返回值: - 返回构建好的Resnext50_32x4d模型实例。如果 `pretrained` 为 `True`, 则返回在ImageNet上预训练的模型。 代码示例: >>> import jittor as jt >>> from jittor.models.resnet import * >>> net = Resnext50_32x4d(pretrained=False) >>> x = jt.rand(1, 3, 224, 224) >>> y = net(x) >>> y.shape [1, 1000] ''' kwargs['groups'] = 32 kwargs['width_per_group'] = 4 model = _resnet(Bottleneck, [3, 4, 6, 3], **kwargs) if pretrained: model.load("jittorhub://resnext50_32x4d.pkl") return model
resnext50_32x4d = Resnext50_32x4d
[文档] def Resnext101_32x8d(pretrained=False, **kwargs): ''' 构建一个Resnext101_32x8d模型 Resnext源自论文 `Aggregated Residual Transformations for Deep Neural Networks <https://arxiv.org/abs/1611.05431>`__。 参数: - `pretrained` (bool, optional): 表示是否预加载预训练模型。默认为 `False`。 返回值: - 返回构建好的Resnext101_32x8d模型实例。如果 `pretrained` 为 `True`, 则返回在ImageNet上预训练的模型。 代码示例: >>> import jittor as jt >>> from jittor.models.resnet import * >>> net = Resnext101_32x8d(pretrained=False) >>> x = jt.rand(1, 3, 224, 224) >>> y = net(x) >>> y.shape [1, 1000] ''' kwargs['groups'] = 32 kwargs['width_per_group'] = 8 model = _resnet(Bottleneck, [3, 4, 23, 3], **kwargs) if pretrained: model.load("jittorhub://resnext101_32x8d.pkl") return model
resnext101_32x8d = Resnext101_32x8d
[文档] def Wide_resnet50_2(pretrained=False, **kwargs): ''' 构建一个Wide_resnet50_2模型 Wide_resnet源自论文 `Wide Residual Networks <https://arxiv.org/abs/1605.07146>`__, 是Resnet的一个变种, 其主要改进是增加了网络的宽度, 即增加了每层的通道数, 从而提高了网络的性能。 参数: - `pretrained` (bool, optional): 表示是否预加载预训练模型。默认为 `False`。 返回值: - 返回构建好的Wide_resnet50_2模型实例。如果 `pretrained` 为 `True`, 则返回在ImageNet上预训练的模型。 代码示例: >>> import jittor as jt >>> from jittor.models.resnet import * >>> net = Wide_resnet50_2(pretrained=False) >>> x = jt.rand(1, 3, 224, 224) >>> y = net(x) >>> y.shape [1, 1000] ''' kwargs['width_per_group'] = (64 * 2) model = _resnet(Bottleneck, [3, 4, 6, 3], **kwargs) if pretrained: model.load("jittorhub://wide_resnet50_2.pkl") return model
wide_resnet50_2 = Wide_resnet50_2
[文档] def Wide_resnet101_2(pretrained=False, **kwargs): ''' 构建一个Wide_resnet101_2模型 Wide_resnet源自论文 `Wide Residual Networks <https://arxiv.org/abs/1605.07146>`__, 是Resnet的一个变种, 其主要改进是增加了网络的宽度, 即增加了每层的通道数, 从而提高了网络的性能。 参数: - `pretrained` (bool, optional): 表示是否预加载预训练模型。默认为 `False`。 返回值: - 返回构建好的Wide_resnet101_2模型实例。如果 `pretrained` 为 `True`, 则返回在ImageNet上预训练的模型。 代码示例: >>> import jittor as jt >>> from jittor.models.resnet import * >>> net = Wide_resnet101_2(pretrained=False) >>> x = jt.rand(1, 3, 224, 224) >>> y = net(x) >>> y.shape [1, 1000] ''' kwargs['width_per_group'] = (64 * 2) model = _resnet(Bottleneck, [3, 4, 23, 3], **kwargs) if pretrained: model.load("jittorhub://wide_resnet101_2.pkl") return model
wide_resnet101_2 = Wide_resnet101_2