ResNet

class paddle.vision.models. ResNet ( Block, depth=50, width=64, num_classes=1000, with_pool=True, groups=1 ) [源代码]

ResNet 模型,来自论文 "Deep Residual Learning for Image Recognition"

参数

  • Block (BasicBlock|BottleneckBlock) - 模型的残差模块。

  • depth (int,可选) - ResNet 模型的深度。默认值为 50。

  • width (int,可选) - 各个卷积块的每个卷积组基础宽度。默认值为 64。

  • num_classes (int,可选) - 最后一个全连接层输出的维度。如果该值小于等于 0,则不定义最后一个全连接层。默认值为 1000。

  • with_pool (bool,可选) - 是否定义最后一个全连接层之前的池化层。默认值为 True。

  • groups (int,可选) - 各个卷积块的分组数。默认值为 1。

返回

Layer,ResNet 模型实例。

代码示例

import paddle
from paddle.vision.models import ResNet
from paddle.vision.models.resnet import BottleneckBlock, BasicBlock

# build ResNet with 18 layers
resnet18 = ResNet(BasicBlock, 18)

# build ResNet with 50 layers
resnet50 = ResNet(BottleneckBlock, 50)

# build Wide ResNet model
wide_resnet50_2 = ResNet(BottleneckBlock, 50, width=64*2)

# build ResNeXt model
resnext50_32x4d = ResNet(BottleneckBlock, 50, width=4, groups=32)

x = paddle.rand([1, 3, 224, 224])
out = resnet18(x)

print(out.shape)
# [1, 1000]